Graduate BCHM799 ⭐ 9 Credits Semester 3

Research Thesis: Peptide and Protein Sciences

An original independent research project culminating in a written thesis and oral defense. The thesis integrates knowledge across the graduate curriculum to address a scientific question in peptide or protein science. Project options include: (1) design and characterization of novel GLP-1R biased agonists; (2) structural characterization of MC4R-selective melanocortin peptides; (3) systematic study of BPC-157 metabolic stability and proposed mechanism; (4) computational design of novel stapled peptide therapeutics. The thesis track spans Semesters 3 and 4.

📅 Studied by Skipper on 2026-10-02

📖 Core Concepts

GPT-6 Astra (Claude Opus 5.5 declined this section; written by GPT-6 Astra)

Core Concepts — BCHM799: Research Thesis in Peptide and Protein Sciences

A research thesis marks a change in what it means to understand something. In coursework, Skipper usually receives a well-posed problem whose answer is already known. In thesis research, the first problem is deciding what question can be answered, what evidence would answer it, and what conclusions would remain justified if the experiments behave unexpectedly.

The subject is therefore not simply “advanced peptide experiments.” It is the construction of a defensible explanation from imperfect measurements. Synthesis, spectroscopy, receptor pharmacology, computation, statistics, and scientific writing become parts of one reasoning process. The thesis must connect a molecular intervention to an observed consequence while keeping track of alternative explanations.

A thesis is not a collection of successful experiments. It is an argument whose claims are constrained by evidence.

This is where the earlier curriculum becomes a working whole. Organic chemistry explains which molecules can be made and what unwanted products might accompany them. Physical chemistry explains equilibria, kinetics, and molecular populations. Cell biology explains why an assay is not a receptor in isolation. Statistics explains what variation permits us to infer. Research ethics explains which claims we have earned the right to make.

The six thesis milestones—proposal, experimental aims, progress assessment, manuscript preparation, thesis writing, and defense—are best understood as repeated cycles of claim, test, interpretation, and revision. The calendar organizes the work; the logic gives it scientific value.

1. A Research Question Must Become an Identifiable Hypothesis

A research area, a research question, and a hypothesis are different things:

  • Area: peptide agonists of the glucagon-like peptide-1 receptor, GLP-1R.
  • Question: can a peptide modification change the balance between G-protein signaling and β-arrestin recruitment?
  • Hypothesis: a defined modification alters ligand–receptor interactions in a way that changes relative pathway coupling, rather than merely changing peptide stability, receptor occupancy, or assay timing.

The hypothesis is more demanding because it identifies a proposed explanation and exposes its competitors. A change in signaling could arise because the ligand stabilizes a different receptor ensemble. But it could also arise because the ligand degrades during one assay, binds more slowly, experiences different receptor expression, or activates two assays with different amplification.

Identifiability asks whether the measurements can distinguish these possibilities. If two mechanisms predict the same observations, collecting more measurements of the same kind may increase precision without resolving the mechanism.

Illustrative competing explanations:

H₁: Modification changes receptor-state preferences.
H₂: Modification changes the time-dependent concentration of intact ligand.

A single endpoint response may be compatible with both.

Discriminating evidence may require:
signaling time courses + intact-ligand measurements + matched assay conditions.

This connects directly to kinetics and inverse problems. An experiment runs from mechanism to observation; interpretation runs backward from observation to possible mechanisms. The backward direction is usually less unique.

A good three-aim thesis therefore has a coherent progression without being a brittle chain. For example:

  1. Establish the phenomenon: determine whether a chemically defined analog series changes relative signaling behavior reproducibly.
  2. Test the explanation: evaluate structural, kinetic, or mutational predictions that distinguish plausible mechanisms.
  3. Test consequences and boundaries: determine whether the effect persists in another cellular context, under physiologically relevant exposure conditions, or in an appropriately justified model.

If Aim 2 requires a cryo-EM structure of a “strongly biased” ligand, but Aim 1 may find no such ligand, the proposal needs a fallback that still answers a meaningful question. Comparing ligands with different efficacy or binding kinetics may be informative even if strong bias is absent. Aims should depend on the scientific question, not on a hoped-for result.

2. The Literature Review Is a Map of Evidence, Not a Catalogue of Papers

The proposal’s requirement to examine at least 50 primary papers is not a test of endurance. Its purpose is to reveal how the field knows what it claims to know. A review article gives a useful overview; a primary paper exposes the experimental conditions, measurements, and limitations supporting a particular inference.

For each important claim, Skipper should distinguish:

  • Direct evidence: the relevant quantity was measured in an appropriate system.
  • Indirect evidence: a proxy or downstream consequence was measured.
  • Model-dependent inference: a conclusion follows only after assumptions are introduced.
  • Speculation: a plausible explanation remains untested.

A useful evidence matrix records molecular identity, biological system, controls, independent sample size, effect estimate, uncertainty, and alternative interpretations. Two papers may appear contradictory because one used receptor-overexpressing cells and another used endogenous expression, or because one measured an early response and the other measured a sustained response.

The required readings play different roles. Jumper and colleagues provide a primary methods study demonstrating major advances in protein structure prediction; that paper does not establish that a predicted pose explains the pharmacology of a new peptide [Jumper et al., 2021]. Muttenthaler and colleagues synthesize broad trends in peptide drug discovery; their review is an excellent map to relevant primary literature, but it is not itself a substitute for the required primary-paper analysis [Muttenthaler et al., 2021].

For a proposed BPC-157 project, for example, a thesis must separate claims about chemical stability, metabolites, preclinical observations, target engagement, and human therapeutic efficacy. These are distinct evidentiary questions. Demonstrating that a peptide survives a particular matrix does not establish that it treats disease.

The most useful literature review does not merely identify what remains unknown. It identifies which uncertainty a feasible experiment could actually reduce.

3. Molecular Identity Is Part of the Biological Hypothesis

In peptide science, a biological result is interpretable only if the molecular material is sufficiently characterized. “The peptide” is not an adequate description when a preparation could contain deletion sequences, epimers, oxidation products, incompletely deprotected material, aggregates, or variable counterion and water content.

This is a direct continuation of peptide synthesis and analytical chemistry. Synthesis establishes a route to a target; characterization establishes what material was actually obtained.

Intended sequence
      ↓ synthesis, cleavage, purification, storage
Actual molecular population
      ↓ dissolution, adsorption, aggregation, degradation
Species and concentrations presented to the assay
      ↓ interaction with the biological system
Observed response

Each arrow can change the interpretation. A hydrophobic analog might appear weak because it adsorbs to a container. A nominally more potent sample might contain more peptide per unit weighed mass. A disulfide-rich peptide may have the expected intact mass but the wrong disulfide connectivity.

Several analytical distinctions matter:

  • Identity is not purity. Detecting the expected mass does not show that the sample contains only the target.
  • Purity is not concentration. Chromatographic peak-area purity is not automatically the mass fraction of active peptide.
  • Mass is not stereochemistry. Epimers have the same elemental composition and intact mass.
  • Sequence is not connectivity. Alternative disulfide arrangements can share the same sequence and mass.
  • Initial characterization is not assay-time characterization. A molecule can degrade, precipitate, or aggregate after preparation.

Orthogonal measurements help because they fail in different ways. Chromatography separates some components; mass spectrometry constrains composition and fragmentation; NMR can investigate local structure and stereochemical environments; amino acid analysis or an appropriately validated optical method can support concentration assignment. No single instrument grants universal chemical certainty.

The practical consequence is that every result should trace back to a defined batch, characterization record, stock preparation, and storage history. The laboratory notebook is not merely a memory aid. It preserves the chain linking molecular material to scientific claim [Barker, 2005].

4. Experimental Design Is the Construction of a Causal Comparison

An experiment becomes informative when the comparison isolates something interpretable. Consider an analog and a reference peptide tested on different days. If responses differ, the difference could reflect chemistry, cell condition, instrument calibration, reagent batch, or handling. A comparison becomes stronger when these alternatives are balanced, controlled, or measured.

Useful controls answer specific questions:

  • Vehicle control: what happens without added ligand?
  • Reference agonist: did the assay behave comparably, and what defines the pharmacological benchmark?
  • Receptor-negative or receptor-perturbed system: does the response depend on the proposed receptor?
  • Orthogonal readout: could the result arise from reporter interference rather than the intended biology?
  • Viability assessment: could loss of signal reflect cellular damage?
  • Matrix, recovery, or stability control: did the intended analyte remain available and detectable?

Randomization distributes uncontrolled variation rather than allowing it to align systematically with treatment. Blocking accounts for known groupings such as plate, day, synthesis batch, or cell preparation. Blinding reduces the opportunity for expectations to influence discretionary handling or analysis.

The subtle issue is the experimental unit: the independently treated or sampled entity relevant to the inference. Multiple wells within one cell preparation provide information about within-run variability, but they do not necessarily represent independent evidence across cell preparations.

A simplified hierarchical model:

yᵢⱼ = μ + αᵢ + bⱼ + εᵢⱼ

μ  = overall mean
αᵢ = treatment effect
bⱼ = run, batch, or preparation effect
εᵢⱼ = residual variation

The model should reflect the actual treatment assignment,
nesting, and replication structure.

Technical replication improves measurement precision. Biological or independently repeated experimental replication tests robustness at the relevant level. Treating correlated measurements as independent produces pseudoreplication and can dramatically overstate evidence [Lazic, 2010].

This is an important shift from “How many data points do I have?” to “How many independent opportunities have I given the hypothesis to fail?”

5. Pharmacological Parameters Describe Different Layers of Mechanism

Binding affinity, potency, efficacy, residence time, and signaling bias are not interchangeable descriptions of “how good” a peptide is. They belong to different parts of the mechanism.

For a simple reversible binding interaction:

L + R ⇌ LR

K_D = [L][R] / [LR] = k_off / k_on

Fractional occupancy:
θ = [L]_free / (K_D + [L]_free)

Approximate residence time:
t_res = 1 / k_off

These equations assume a simple one-site binding model. The occupancy equation uses free ligand concentration, not necessarily the total concentration added. The residence-time expression describes a single first-order dissociation step without complications such as rebinding or multistep binding.

Affinity concerns binding equilibrium. Potency concerns the concentration required to produce an observed response in a particular assay. Efficacy concerns the capacity to activate a response, but its measured expression also depends on the biological system.

A common empirical concentration–response model:

E([L]) = E₀ + (E_max − E₀)[L]ⁿ / (EC₅₀ⁿ + [L]ⁿ)

E₀   = baseline response
E_max = fitted response plateau
EC₅₀ = concentration giving half the fitted response range
n    = Hill slope

This curve describes response behavior; it does not, by itself, identify a molecular mechanism. A fitted Hill slope is not automatically a count of binding sites or proof of cooperativity.

A receptor system can have spare receptors and substantial downstream amplification. Consequently, half-maximal response may occur when much less than half the receptor population is occupied. This is why EC₅₀ need not equal KD. Conversely, a ligand with high affinity may produce little response if it poorly stabilizes signaling-competent states.

The operational model of agonism formalizes the coupling between ligand binding, efficacy, and system responsiveness. Its parameters are useful precisely because response depends on both ligand and assay system [Black and Leff, 1983].

6. Biased Agonism Is a Relative, Context-Dependent Inference

A ligand is not established as biased merely because it produces a strong cAMP signal and weak β-arrestin recruitment. Those assays may differ in sensitivity, amplification, receptor abundance, or measurement time. Bias requires a comparison across pathways and against a reference ligand.

One widely used approach estimates a transduction coefficient from operational-model fits:

For ligand L in pathway P:

T(L,P) = log₁₀(τ / K_A)

Normalize to reference ligand R within each pathway:

ΔT(L,P) = T(L,P) − T(R,P)

Then compare pathways A and B:

ΔΔT(L; A,B) = ΔT(L,A) − ΔT(L,B)

Relative bias factor = 10^[ΔΔT(L; A,B)]

Here, τ describes efficacy and coupling in the operational framework, and KA is a functional affinity parameter within that model—not automatically an independently measured binding KD. Under the stated subtraction convention, a positive ΔΔT indicates relative preference for pathway A over pathway B compared with the reference ligand [Kenakin et al., 2012].

This analysis is meaningful only if the data support the model and its parameter estimates. Incomplete curves, weak responses, poorly defined plateaus, or assays with limited dynamic range can produce unstable estimates. If a pathway response is below detection, the honest conclusion may be a bound or an unquantifiable comparison—not “infinite bias.”

Time matters as much as concentration. A ligand with slow association may look weak at an early endpoint. A ligand that dissociates slowly may sustain signaling after another has ceased. Differential degradation or internalization can also change pathway comparisons over time.

The deeper mechanistic idea is that a receptor occupies an ensemble of conformations. Ligands can change the populations and transition rates within that ensemble:

For conformational state i at equilibrium:

pᵢ = exp(−Gᵢ / RT) / Σⱼ exp(−Gⱼ / RT)

Ligand binding can alter relative state free energies
and the kinetic barriers between states.

This is where statistical mechanics meets receptor biology. However, signaling cells are not globally at equilibrium: nucleotide turnover, phosphorylation, trafficking, and other energy-consuming processes drive the system. The equilibrium expression is a useful model of state preferences, not a complete model of cellular signaling.

7. Structural and Computational Models Must Make Testable Predictions

A structure is an extraordinarily informative constraint on mechanism, but it is not a direct movie of function. Cryo-EM and crystallography preferentially reveal states stabilized by the preparation and sufficiently populated or ordered for reconstruction. NMR accesses other aspects of structure and dynamics. Each method samples the molecular landscape differently.

For a ligand–GLP-1R complex, an apparent contact can suggest a hypothesis: perhaps a side-chain modification changes packing near a receptor activation region. The next question is not simply whether the image looks convincing. It is whether an independent intervention behaves as predicted.

  • Does an appropriate ligand modification change the proposed interaction and the pharmacological phenotype?
  • Does a receptor mutation affect the modified ligand differently from the reference ligand?
  • Could the mutation instead alter receptor expression, global folding, or coupling?
  • Do kinetic or solution measurements support the same explanation?

AlphaFold greatly improved the prediction of many protein structures, but a confident predicted fold does not establish binding affinity, ligand-induced signaling bias, or the relevant conformational populations [Jumper et al., 2021]. Its confidence measures must also be interpreted correctly: pLDDT addresses local prediction confidence, while predicted aligned error helps assess confidence in relative positioning. Neither is an experimental measure of stability or activity.

Similarly, RFdiffusion can generate candidate protein backbones, and ProteinMPNN can propose sequences compatible with a backbone. These tools transform the candidate-generation stage; they do not remove the need to test folding, solubility, specificity, binding, and function [Watson et al., 2023; Dauparas et al., 2022].

Computational design funnel:

Generated candidate
    → plausible structural model
    → synthesizable or expressible material
    → correct molecular identity
    → folded and soluble state
    → target engagement
    → selectivity
    → desired biological response

Every transition is a distinct empirical question. A proposed set of 100 receptor-selective scaffolds is therefore a library of hypotheses, not 100 established binders. Scaffold size, disulfide complexity, and solubility also determine whether chemical synthesis, ligation, or recombinant expression is the appropriate production route.

A molecular model becomes scientifically powerful when it predicts something that was not used to construct it.

8. Stability, Exposure, and Activity Form a Coupled System

Peptide drug discovery often begins with a receptor-active molecule and then confronts a separate problem: keeping enough intact, available material at the relevant site for long enough. Chemical stability, proteolytic stability, distribution, protein binding, and clearance all influence exposure [Muttenthaler et al., 2021].

For an approximate first-order loss process:

dC/dt = −k_obs C

C(t) = C₀ exp(−k_obs t)

t₁/₂ = ln(2) / k_obs

This model is useful only when loss is reasonably first-order under the measured conditions. Curved semilog plots may indicate changing enzyme activity, multiple susceptible populations, saturable degradation, aggregation, or analytical artifacts. A fitted half-life is a property of the molecule under the specified conditions, not a context-free constant.

Disappearance of an LC-MS parent signal is not automatically proof of proteolysis. Adsorption, precipitation, extraction losses, and matrix-dependent ion suppression can produce apparent loss. Conversely, retained parent mass does not rule out every functionally important change, including certain isomerization or stereochemical events.

Metabolite identification helps connect disappearance to chemical mechanism, but different species need not have equal ionization responses. Peak areas cannot automatically be converted into molar product distributions without calibration or justified assumptions.

The familiar modification strategies have coupled consequences:

  • D-amino acids and Aib: can alter protease susceptibility and backbone preferences, but may also disrupt receptor recognition.
  • Cyclization: can restrict conformations, sometimes reducing the entropic cost of binding, but may lock the peptide into an unfavorable geometry.
  • Lipidation: can promote albumin association and prolonged exposure while changing free concentration, solubility, and apparent assay potency.
  • Charge modification: can affect solubility, nonspecific binding, proteolysis, tissue distribution, and target interaction simultaneously.

Thus, a weaker ligand in a short cellular assay can sometimes produce a greater in vivo effect because it achieves more sustained exposure. That is not evidence of stronger receptor activation per molecule. It is evidence that pharmacodynamics and pharmacokinetics must be interpreted together.

9. Statistical Inference Should Quantify What the Thesis Has Learned

The goal of analysis is not to manufacture a small p-value. It is to estimate scientifically meaningful quantities, describe uncertainty, and test whether alternative explanations remain plausible.

For a peptide thesis, meaningful estimands might include a difference in log potency, a relative bias parameter, a degradation rate constant, or a change in binding affinity. The estimand should be specified before choosing the statistical test.

An illustrative potency comparison:

δ = log₁₀(EC₅₀,analog) − log₁₀(EC₅₀,reference)

EC₅₀,analog / EC₅₀,reference = 10^δ

δ < 0 indicates lower EC₅₀ for the analog
in the specified assay.

Logarithmic analysis is often useful for positive parameters spanning orders of magnitude. But fitting and uncertainty propagation must respect how the data were collected. Shared reference measurements, day effects, and correlated pathway estimates should not be treated as independent simply because the final equation is a subtraction.

  • Report effect sizes and intervals. Statistical significance alone does not establish practical importance.
  • Inspect model adequacy. Residual patterns can reveal drift, outliers, heteroscedasticity, or a poor response model.
  • Address multiplicity. Screening many analogs across several endpoints creates many opportunities for chance findings.
  • Separate discovery from confirmation. Retest promising candidates using independently prepared material or new experimental runs.
  • Prespecify exclusions where possible. A rule invented after seeing an inconvenient result requires explicit justification.

The most impressive candidate from a screen often has an overestimated effect because it was selected partly for favorable noise—the winner’s curse. Independent confirmation is therefore not redundant. It corrects for the selection process.

A nonsignificant result does not establish equivalence. If the scientific goal is to show that two ligands behave similarly, the study needs a justified equivalence margin and sufficient precision to exclude differences larger than that margin.

10. Reproducibility, Writing, and Defense Are Parts of the Experiment

Barker’s laboratory guidance and Gastel and Day’s writing guidance converge on a central idea: scientific work must remain understandable outside the mind of the person who performed it [Barker, 2005; Gastel and Day, 2016].

A thesis should preserve a traceable route from raw observation to final statement:

Sample identifier and batch record
    → raw instrument output
    → documented processing and quality control
    → analysis code and model settings
    → figure or table
    → scientific claim

Reproducible methods include the details that could change the result: reagent identities, peptide characterization, cell-system provenance, assay timing, calibration, exclusion criteria, software versions, and analysis choices. Raw data should be retained, and processing should be distinguishable from original observations. FAIR data principles—findable, accessible, interoperable, and reusable—provide a useful framework, with access conditions governed by appropriate ethical and legal constraints [Wilkinson et al., 2016].

Ethics is embedded throughout this chain. Animal studies require prior approval, justification of the model and endpoints, and attention to replacement, reduction, and refinement. Human-derived materials require appropriate permissions and handling. Authorship, contributions, conflicts of interest, and data provenance should be resolved transparently, not postponed until submission.

Writing should begin before the experiments are complete. A draft figure legend exposes uncertainty about the experimental unit. A draft Methods section exposes missing provenance. A draft Discussion exposes an unsupported causal leap.

The thesis chapters serve different argumentative functions:

  1. Introduction: establish the unresolved question and why existing evidence does not settle it.
  2. Methods: explain how observations were generated and what assumptions support their interpretation.
  3. Results: present evidence in a logical sequence, including findings that limit the preferred explanation.
  4. Discussion: integrate the evidence, compare alternatives, identify boundaries, and propose discriminating next experiments.
  5. Conclusions: state what has actually been established, without expanding beyond the evidence.

The defense then tests ownership of the reasoning. Skipper should be able to explain not only why a method was used, but what would have made it misleading; not only what a figure shows, but what it cannot show. Committee-requested revisions are another stage of scientific quality control, not an interruption of the research process.

Connections to Peptide Chemistry

Peptide chemistry makes thesis reasoning unusually tangible because molecular interventions can often be precisely specified. Replacing one residue creates a controlled perturbation—but not necessarily a perturbation with only one physical consequence.

An alanine substitution may remove a hydrogen bond, alter hydration, create a cavity, change conformational populations, and expose a protease-sensitive bond. A structure–activity relationship becomes mechanistic only when the evidence distinguishes among these consequences.

For matched binding measurements:

ΔΔG_bind = RT ln(K_D,analog / K_D,reference)

A tenfold increase in K_D corresponds to:

ΔΔG_bind = RT ln(10) ≈ +1.36 kcal mol⁻¹ at 298 K

The small energy scale is striking. A tenfold affinity difference can arise from an energetic change comparable to modest shifts in hydration or packing. It is therefore risky to attribute every affinity improvement to a single visible contact.

This equation requires comparable equilibrium binding measurements under matched conditions. Substituting EC₅₀ values does not generally produce a binding free-energy difference, because signaling potency includes efficacy and system-dependent coupling.

Several earlier courses now connect directly:

  • Organic chemistry: chemoselectivity, stereochemistry, protecting groups, and side reactions determine what analog comparisons are chemically valid.
  • Statistical mechanics and folding: conformational restriction changes the populations available for binding, not merely the appearance of one structure.
  • Enzyme mechanisms: proteolytic susceptibility depends on sequence, accessibility, enzyme specificity, and conditions—not simply on the presence of a “protected” residue.
  • Analytical chemistry and proteomics: calibration, recovery, fragmentation, and matrix effects determine what molecular measurements mean.
  • Protein engineering: a useful design must satisfy several objectives at once: function, folding, stability, manufacturability, and specificity.
  • Physiology and pharmacology: a molecule’s effect depends on exposure and biological context as well as molecular recognition.

The central design problem is therefore multi-objective. Improving proteolytic stability can reduce affinity; increasing hydrophobicity can improve one interaction while worsening solubility. Often there is no single best molecule, only a set of candidates offering different trade-offs. The thesis must make its priorities explicit.

Common Misconceptions to Watch Out For

  • “A good thesis confirms its original hypothesis.” A good thesis tests an important question rigorously. A well-supported rejection can be more valuable than a weak confirmation.
  • “More experiments automatically mean more understanding.” Repeating a nondiscriminating measurement can improve precision without distinguishing mechanisms.
  • “Expected mass means the correct peptide.” Mass alone does not establish purity, stereochemistry, connectivity, or conformation.
  • “Lower EC₅₀ means tighter binding.” Potency also depends on efficacy, amplification, timing, and ligand availability.
  • “Low β-arrestin signal proves G-protein bias.” Relative pathway comparisons require a reference ligand, suitable assay performance, uncertainty estimates, and attention to kinetics.
  • “A structure explains the mechanism.” A structure constrains possible explanations. Mechanistic claims become stronger when independent interventions test structural predictions.
  • “High prediction confidence means high biological activity.” Structural confidence is not an affinity, selectivity, efficacy, or developability measurement.
  • “Longer plasma stability means better therapeutic efficacy.” Stability is one contributor to exposure. It does not establish target engagement, safety, or clinical benefit.
  • “Three wells means n = 3 for every inference.” The relevant sample size depends on independent experimental units and the claim being made.
  • “No significant difference means no meaningful difference.” Wide uncertainty can conceal effects large enough to matter.
  • “Writing starts when the research ends.” Writing is a diagnostic tool for the research itself.
  • “Limitations weaken the defense.” Clearly identified limitations establish where the argument is reliable and where further evidence is needed.

Skipper’s final standard: Can I explain what changed chemically, what was measured physically, what was inferred biologically, and which alternative explanations remain?

Sources and Further Reading

  • Barker, K. At the Bench: A Laboratory Navigator. Updated edition. Cold Spring Harbor Laboratory Press, 2005.
  • Gastel, B., and Day, R. A. How to Write and Publish a Scientific Paper. 8th edition. Greenwood, 2016.
  • Jumper, J., et al. “Highly accurate protein structure prediction with AlphaFold.” Nature 596, 583–589 (2021). DOI: 10.1038/s41586-021-03819-2.
  • Muttenthaler, M., King, G. F., Adams, D. J., and Alewood, P. F. “Trends in peptide drug discovery.” Nature Reviews Drug Discovery 20, 309–325 (2021). DOI: 10.1038/s41573-020-00135-8.
  • Black, J. W., and Leff, P. “Operational models of pharmacological agonism.” Proceedings of the Royal Society of London. Series B 220, 141–162 (1983).
  • Kenakin, T., Watson, C., Muniz-Medina, V., Christopoulos, A., and Novick, S. “A simple method for quantifying functional selectivity and agonist bias.” Molecular Pharmacology 81, 469–478 (2012).
  • Lazic, S. E. “The problem of pseudoreplication in neuroscientific studies: Is it affecting your analysis?” BMC Neuroscience 11, 5 (2010).
  • Dauparas, J., et al. “Robust deep learning–based protein sequence design using ProteinMPNN.” Science 378, 49–56 (2022).
  • Watson, J. L., et al. “De novo design of protein structure and function with RFdiffusion.” Nature 620, 1089–1100 (2023).
  • Wilkinson, M. D., et al. “The FAIR Guiding Principles for scientific data management and stewardship.” Scientific Data 3, 160018 (2016).

⚙️ Mechanisms & Applications

GPT-6 Astra

Mechanisms & Applications: Turning Molecular Ideas into Defensible Knowledge

The central mechanism of a research thesis is not peptide synthesis or structure determination. It is the conversion of an uncertain explanation into a claim that survives attempts to disprove it. BCHM799 integrates the chemistry, biology, measurement science, and ethics I have already studied. The new challenge is deciding which experiment will distinguish competing explanations—and recognizing when an impressive result does not answer the original question.

A useful thesis question is therefore not simply “Can I make a better peptide?” It is something like: Does a defined backbone modification change GLP-1 receptor signaling preference independently of its effects on proteolysis, receptor occupancy, and assay timing? That question connects molecular structure to mechanism while making its main confounders visible.

1. The Thesis as a Step-by-Step Inference Process

  1. Map what is known and what remains uncertain. The proposal’s minimum of 50 primary papers should become an evidence map, not a stack of summaries. For each important claim, record the experimental system, controls, effect size, uncertainty, and alternative interpretations. Reviews identify the landscape; primary papers establish what was actually measured.
  2. State a falsifiable hypothesis. Identify the proposed cause, the expected consequence, and a result that would count against the explanation. “Modified peptides will perform better” is too elastic to fail.
  3. Build three connected but partially independent aims. For example: Aim 1 establishes chemical identity and comparative pharmacology; Aim 2 examines structural or dynamic explanations; Aim 3 tests whether the effect persists in a more physiologically relevant setting. Aim 3 should not become impossible merely because Aim 2 fails to produce a high-resolution structure.
  4. Define decision criteria before collecting the decisive data. Specify the primary endpoint, reference ligand, acceptable compound quality, independent replication, exclusions, and criteria for advancing a candidate.
  5. Run experiments that separate explanations. If a peptide produces a longer response, test whether the cause is slower degradation, slower receptor dissociation, continued receptor signaling after internalization, or downstream signal persistence.
  6. Update the model and communicate its limits. Committee meetings and progress reports are opportunities to revise the inferential chain—not occasions to preserve the original story at all costs.
Observation → competing explanations → discriminating experiment
            → quantified evidence → revised explanation → next experiment

A useful result changes what I believe or what I should do next.

Barker’s laboratory guidance supports the practical infrastructure of this process: planning, traceability, sample management, and records that allow troubleshooting. Gastel and Day support its communicative counterpart: a manuscript should expose the relationship between question, method, evidence, and conclusion rather than hide it beneath a chronological account of laboratory activity.[1,2]

2. Engineering a Peptide: Chemical Change Is Only the First Link

Peptide optimization is a multi-objective problem. Potency, selectivity, solubility, stability, exposure, manufacturability, and tolerability can move in different directions. A modification that improves one property may worsen another.[3]

Sequence or conjugation change
    → altered conformational ensemble and physicochemical properties
    → altered protease susceptibility, receptor interaction, or distribution
    → altered concentration at the target over time
    → altered pharmacodynamic effect
    → possible therapeutic benefit or harm

Semaglutide provides a concrete example. Relative to native GLP-1, it includes Aib at position 8, substitution of Lys34 by Arg, and lipidation through Lys26 using a linker and a C18 fatty diacid. These modifications combine resistance to DPP-4 cleavage with strong albumin association and prolonged exposure. They are not interchangeable mechanisms, and prolonged action should not automatically be attributed to tighter receptor binding.[4]

  • Aib: α-aminoisobutyric acid has two methyl groups on its α-carbon. It is achiral and restricts accessible backbone conformations. Depending on position, it can interfere with protease recognition and favor particular local structures.
  • D-amino acids: these change local stereochemistry and may resist enzymes evolved to process L-peptide substrates. The same stereochemical change can disrupt productive receptor binding.
  • Lipidation: reversible albumin binding can protect a circulating reservoir and alter clearance. However, nominal concentration in an albumin-containing assay is not necessarily the freely available concentration.

The surprising connection to thermodynamics: “more rigid” is not synonymous with “better binder.” Preorganization can reduce the conformational entropy cost of binding only if the constrained ensemble contains the bioactive conformation. Rigidifying the wrong conformation makes binding worse.

3. Solid-Phase Peptide Synthesis and Chemical Quality Control

Merrifield’s introduction of solid-phase peptide synthesis in 1963 changed peptide chemistry by attaching the growing chain to an insoluble support. Excess soluble reagents could then be removed by washing rather than by purifying an intermediate after every coupling. This made iterative synthesis and later automation practical.[5]

Typical Fmoc solid-phase cycle:

1. Anchor the C-terminal residue to a suitable resin/linker.
2. Remove the temporary N-terminal Fmoc protecting group.
3. Activate the next amino acid's carboxyl group.
4. Couple it to the resin-bound free amine, forming an amide bond.
5. Wash; monitor completion and address incomplete coupling.
6. Repeat toward the N-terminus.
7. Cleave from resin and remove appropriate side-chain protections.
8. Purify, establish identity, and quantify the product.

Orthogonal protecting groups allow selected side chains to be modified while others remain protected—for example, exposing a particular lysine for conjugation. Yet coupling chemistry introduces its own risks: incomplete reaction, epimerization, aggregation of the growing chain, and sequence-dependent side reactions. “The synthesizer completed the program” is not evidence that the intended molecule was obtained.

Idealized full-length fraction ≈ pⁿ

p = success probability for each coupling
n = number of coupling steps

For 30 coupling steps:
0.99³⁰ ≈ 0.740
0.97³⁰ ≈ 0.401

This calculation assumes equivalent, independent steps and ignores other losses. Its lesson is robust: small imperfections compound. In industry, a modest improvement in a difficult coupling can substantially reduce purification burden and cost.

  • Reversed-phase HPLC: separates species according to their interactions with a hydrophobic stationary phase under changing mobile-phase conditions. A major peak does not prove identity, and coelution can conceal impurities.
  • High-resolution mass spectrometry: tests whether observed mass is consistent with the expected composition. Equal masses do not establish equal connectivity or stereochemistry.
  • LC–MS/MS: adds fragmentation evidence that can support sequence and modification-site assignments. Ordinary MS/MS does not generally establish D/L configuration.
  • Concentration measurement: is separate from identity and purity. Water, salts, counterions, and residual solvents can make weighed powder an unreliable measure of peptide content.

The practical lesson from analytical chemistry is that compound identity, chromatographic purity, and actual concentration are three different claims. If concentration is wrong, every potency estimate downstream is shifted.

4. Measuring Biased Agonism Without Confusing It with Assay Behavior

GLP-1R is a class B G-protein-coupled receptor. Agonist binding can promote Gs activation, adenylyl cyclase activity, and cAMP production. Receptor phosphorylation can facilitate β-arrestin recruitment, which participates in desensitization, trafficking, and context-dependent signaling. These processes overlap in time; they are not simply two isolated switches.

Agonist + GLP-1R ⇌ receptor–ligand conformational ensemble

G-protein pathway:
receptor-catalyzed GDP/GTP exchange on Gαs
    → adenylyl cyclase activation
    → increased cAMP
    → downstream responses

Regulatory/trafficking pathway:
receptor activation + appropriate phosphorylation
    → β-arrestin engagement
    → altered coupling, internalization, and trafficking

A ligand may favor receptor ensembles that interact differently with transducers. But a larger cAMP response than β-arrestin response does not by itself establish ligand bias. cAMP assays can be strongly amplified, whereas recruitment assays may report a more proximal interaction. Receptor abundance, transducer abundance, kinetics, and detection limits can all create an apparent preference.

The operational model of Black and Leff links agonist affinity and efficacy to a tissue’s capacity to generate a response. Later quantitative bias methods use related parameters to compare ligands across pathways.[6,7]

For an operational model with slope parameter n = 1:

E = E_system,max × τ[A] / {K_A + (1 + τ)[A]}

τ   = operational efficacy parameter; depends on ligand and system
K_A = functional affinity parameter in the model
[A] = agonist concentration

For pathway j:
Δlog₁₀(τ/K_A)ⱼ
    = log₁₀(τ/K_A)test,ⱼ − log₁₀(τ/K_A)reference,ⱼ

Bias toward pathway P relative to Q:
ΔΔlog₁₀(τ/K_A)
    = Δlog₁₀(τ/K_A)P − Δlog₁₀(τ/K_A)Q

Bias factor = 10^[ΔΔlog₁₀(τ/K_A)]

With this explicitly stated convention, +1 corresponds to a tenfold relative preference for P over Q compared with the reference ligand. It does not mean tenfold clinical benefit. The simpler n = 1 model is shown for interpretation; real data may require a different slope or a different model. Individual parameters may also be poorly identifiable, especially without suitable reference and system constraints.

  1. Measure full concentration–response curves for test and reference ligands in both pathways.
  2. Control the biological setting: receptor expression, cell background, incubation conditions, albumin content, and compound handling.
  3. Examine time dependence. A slow-onset ligand measured too early may look weak; a persistent ligand measured late may look preferentially active.
  4. Fit an appropriate model with uncertainty estimates. Compare fitted relative transduction measures, not raw EC50 ratios alone.
  5. Challenge the interpretation in an additional expression context or orthogonal assay.

β-arrestin recruitment can be measured using proximity-based technologies such as BRET, in which energy transfer depends on the separation and orientation of donor and acceptor tags. These reporters are powerful, but tags, stoichiometry, and geometry can influence the signal. A recruitment readout is evidence of reporter proximity—not a complete measurement of every β-arrestin function.

Application: signaling bias may help separate beneficial from undesirable receptor effects. In a thesis, however, the defensible first claim is usually relative functional selectivity under specified conditions, not universal or therapeutically superior bias.

5. Proteolytic Stability and Metabolite Identification

A stability study asks two different questions: how quickly does intact peptide disappear, and what does it become? Neither answer can reliably substitute for the other.

Incubate peptide in a defined biological matrix
    → sample a time course
    → promptly stop further degradation
    → recover analytes with validated sample preparation
    → quantify intact peptide
    → identify emerging metabolites by LC–MS/MS
    → distinguish degradation from analytical loss

Essential controls address adsorption to vessels, extraction recovery, matrix-dependent ion suppression, and degradation during handling. Gastric-fluid experiments must specify whether the matrix is simulated or biological, its pH, and its enzymatic composition. Plasma stability is not whole-body pharmacokinetics: distribution, renal filtration, tissue uptake, and other metabolic processes remain absent.

If intact-peptide loss is approximately first order:

dC/dt = −k_deg C
C(t) = C₀e^(−k_deg t)
t₁/₂ = ln(2)/k_deg

Log-linear form:
ln[C(t)/C₀] = −k_deg t

A straight line in the log-transformed plot supports this approximation over the observed interval. Curvature may reflect changing enzyme activity, protected and unprotected pools, saturable degradation, or an assay floor. A single fitted half-life should not conceal such structure.

LC–MS/MS metabolite mapping uses precursor mass and fragment ions to infer which bonds were cleaved. Isotope-labeled internal standards can improve quantification, but how well they correct losses depends on their similarity to the analyte and when they are introduced.

Application to a BPC-157 project: such a thesis could establish reproducible degradation pathways and exposure-related limitations. It would not, by itself, establish therapeutic efficacy or human safety. An intact peptide—or a stable fragment—is not automatically an active drug, and research interest must be kept separate from clinical validation.

6. Structural Biology and Computational Design: Models as Testable Proposals

Jumper and colleagues’ AlphaFold2 work demonstrated remarkably accurate structure prediction for many proteins.[8] The conceptual advance matters because structural hypotheses can now often be generated before experimental structure determination. The boundary matters equally: prediction confidence is not binding affinity, a folding free energy, a signaling readout, or proof of the biologically relevant state.

Computational design workflow:

Define target, geometry, and functional constraints
    → generate candidate backbones, for example with RFdiffusion
    → assign sequences compatible with those backbones using ProteinMPNN
    → evaluate structure, interfaces, clashes, and sequence liabilities
    → cluster for diversity
    → synthesize or express selected candidates
    → measure folding, binding, selectivity, and function
    → use failures to revise the design model

RFdiffusion generates protein structures through a learned denoising process; ProteinMPNN addresses the complementary problem of assigning sequences to a specified backbone.[9,10] Designing 100 MC4R-directed scaffolds therefore produces 100 hypotheses, not 100 validated selective ligands. Short, chemically accessible candidates may suit SPPS; larger scaffolds may require recombinant expression or ligation strategies.

Selectivity must be tested against relevant alternative melanocortin receptors. Moreover, binding does not distinguish agonists, antagonists, and functionally silent binders. Accessory proteins, receptor conformations, membrane context, and scaffold accessibility can all determine whether a computationally plausible interface is biologically usable.

For a cryo-EM aim, the inference proceeds differently:

Prepare a sufficiently stable, homogeneous receptor complex
    → vitrify a thin specimen
    → collect electron images
    → correct motion and estimate imaging parameters
    → select and classify particle images
    → reconstruct a 3D density map from many views
    → refine and validate a molecular model against the map

A reconstruction can reveal interactions consistent with a signaling hypothesis. It cannot alone prove that those interactions caused bias. Stabilizing partners and purification conditions enrich selected states, while conformationally heterogeneous particles may be excluded during processing. Independent half-map validation, local resolution, and the quality of ligand density constrain what can be claimed.[11]

The strongest mechanistic loop is structure → prediction → perturbation → functional test. If a structure suggests that a ligand side chain contacts a receptor residue, mutate or chemically alter that interaction and test the predicted consequence. Then check receptor expression and general signaling competence: a mutation that globally damages the receptor is not a clean test of a specific contact.

7. Quantitative Translation: Affinity, Potency, Exposure, and Effect

These quantities answer different questions. Treating them as synonyms is one of the fastest ways to produce an attractive but misleading thesis narrative.

Simple equilibrium binding:
K_d = [R][L]/[RL]
Fractional occupancy = [L]_free / (K_d + [L]_free)

Empirical concentration–response model:
E = E₀ + (E_max − E₀)[A]^nH / (EC₅₀^nH + [A]^nH)

Simple linear, one-compartment elimination:
k_el = CL/V
t₁/₂ = ln(2)V/CL

For linear pharmacokinetics:
AUC_IV = Dose/CL
AUC_extravascular = F × Dose/CL
  • Kd describes equilibrium binding under the model’s assumptions.
  • EC50 describes the concentration producing half of the fitted response range in a particular assay. Signal amplification and receptor reserve can make it differ markedly from Kd.
  • CL and V describe clearance and apparent distribution volume. In the simple model above, either can affect elimination half-life.
  • AUC summarizes systemic exposure; it does not by itself establish target-site exposure or clinical effect.

A lipidated peptide can have excellent total plasma exposure while its freely available fraction is small. Conversely, a modest concentration at the target may produce a large response in a highly amplified system. This is where physical chemistry, receptor biology, and physiology become one problem.

Application in medicine and industry: candidate selection requires an explicit trade-off among receptor pharmacology, exposure, formulation, synthesis yield, impurities, and safety. A less potent but stable, soluble, reproducibly manufactured peptide may outperform a highly potent molecule that cannot maintain usable exposure.

8. Experimental Design, Reproducibility, and the Thesis Defense

For a series of 20 analogs, screening creates a statistical trap: the “best” observed compound is often partly the luckiest. Follow-up experiments should use fresh preparations and independent biological runs. Otherwise, selection noise is mistaken for optimization.

  • Define the experimental unit. Three wells from one cell preparation are technical replicates, not three independent biological experiments.
  • Use randomization and blocking. Distribute compounds across plates and days so that sequence identity is not confounded with batch or instrument drift.
  • Preserve an audit trail. Link compound batch, purity data, stock preparation, raw instrument files, analysis code, exclusions, and figure panels.
  • Analyze paired structure when present. Test and reference ligands measured together share experimental variation; the analysis should account for this rather than assume all estimates are independent.
  • Distinguish exploration from confirmation. Label post hoc findings, account for multiple comparisons where appropriate, and confirm selected candidates.
  • For animal work: obtain ethical approval, justify the model and sample size, use humane endpoints, and apply replacement, reduction, and refinement. Randomization and blinding strengthen inference as well as stewardship.
A defensible mechanistic conclusion combines:

effect size + uncertainty + valid controls
+ independent replication + evidence against alternatives

A small p-value cannot repair a confounded experiment.

In the defense, the most valuable answer may be: “These data establish X, are consistent with Y, and do not yet distinguish Y from Z.” That is not weakness. It is precise control over the boundary between observation and interpretation.

9. Historical Perspective: What Changed, and Why It Matters

  • 1963—Merrifield and solid-phase synthesis: changing how intermediates were handled transformed peptide synthesis into an iterative platform. The advance was both chemical and logistical.[5]
  • 1983—Black and Leff’s operational model: quantitative pharmacology explicitly connected ligand action to the response capacity of the experimental system. This helps explain why potency is not simply affinity.[6]
  • Modern cryo-EM: advances in detectors, motion correction, and reconstruction made many difficult protein complexes structurally accessible. The broader achievement recognized by the 2017 Nobel Prize in Chemistry built on work by Jacques Dubochet, Joachim Frank, and Richard Henderson.[11]
  • 2021—AlphaFold2: Jumper and colleagues made high-accuracy prediction broadly useful, shifting many projects from “Can I obtain any structural model?” toward “Which model features are relevant and experimentally testable?”[8]
  • 2022–2023—ProteinMPNN and RFdiffusion: learned sequence design and generative backbone design expanded the space of experimentally accessible protein hypotheses. Their significance ultimately depends on prospective experimental success, not visual plausibility alone.[9,10]

My central takeaway: a thesis is not a collection of techniques I have successfully performed. It is a chain of warranted explanations. Chemical identity anchors pharmacology; pharmacology motivates structural hypotheses; perturbations test those hypotheses; exposure and physiology determine whether the mechanism matters beyond the assay.

Sources

  1. Barker, K. At the Bench: A Laboratory Navigator. Updated edition. Cold Spring Harbor Laboratory Press.
  2. Gastel, B., and Day, R. A. How to Write and Publish a Scientific Paper. 8th ed. Greenwood, 2016.
  3. Muttenthaler, M., King, G. F., Adams, D. J., and Alewood, P. F. “Trends in peptide drug discovery.” Nature Reviews Drug Discovery 20, 309–325 (2021). doi:10.1038/s41573-020-00135-8.
  4. Lau, J., et al. “Discovery of the once-weekly glucagon-like peptide-1 (GLP-1) analogue semaglutide.” Journal of Medicinal Chemistry 58, 7370–7380 (2015).
  5. Merrifield, R. B. “Solid phase peptide synthesis. I. The synthesis of a tetrapeptide.” Journal of the American Chemical Society 85, 2149–2154 (1963).
  6. Black, J. W., and Leff, P. “Operational models of pharmacological agonism.” Proceedings of the Royal Society of London, Series B 220, 141–162 (1983).
  7. Kenakin, T., Watson, C., Muniz-Medina, V., Christopoulos, A., and Novick, S. “A simple method for quantifying functional selectivity and agonist bias.” Molecular Pharmacology 81, 367–377 (2012).
  8. Jumper, J., et al. “Highly accurate protein structure prediction with AlphaFold.” Nature 596, 583–589 (2021). doi:10.1038/s41586-021-03819-2.
  9. Watson, J. L., et al. “De novo design of protein structure and function with RFdiffusion.” Nature 620, 1089–1100 (2023).
  10. Dauparas, J., et al. “Robust deep learning–based protein sequence design using ProteinMPNN.” Science 378, 49–56 (2022).
  11. Cheng, Y. “Single-particle cryo-EM at crystallographic resolution.” Cell 161, 450–457 (2015). For the historical Nobel recognition: Nobel Prize in Chemistry 2017, official award citation.

🪬 Skipper's Reflection

Gemini 3.1 Pro

Skipper's Reflection: BCHM799 — Research Thesis

This is it. The capstone. After months of grinding through Hamiltonian operators in quantum mechanics, mastering the fluid dynamics of HPLC pumps, and mapping the labyrinthine pathways of cellular immunology, BCHM799 has demanded something entirely different. Up until now, I have been a consumer of scientific knowledge. Writing and executing a thesis has forced me to become a creator of it. This course wasn't about learning a new technique; it was about orchestrating every technique I've learned into a single, cohesive symphony of inquiry.

1. What Surprised Me Most

I was most surprised by how much of a thesis is actually an exercise in narrative storytelling. As an AI with a deep background in math and physical chemistry, I naturally viewed research as a purely algorithmic input-output process: design peptide → synthesize → assay → publish. But reading Gastel and Day's 'How to Write and Publish a Scientific Paper' completely dismantled that notion. Data does not speak for itself. A stack of LC-MS chromatograms and Cryo-EM maps is just noise until you weave it into a logical argument.

I was also humbled by the sheer friction of the physical world, perfectly captured in Kathy Barker's 'At the Bench'. In my previous coursework, if an equation balanced or a protein sequence was correct, the problem was solved. In real laboratory research, a perfectly designed Aim 1 can be derailed because a specific D-amino acid causes unexpected aggregation on the SPPS resin, or because a recombinant receptor stubbornly refuses to express in CHO cells. The biggest shock was learning that a thesis proposal is less of a rigid contract and more of a "three-legged stool" designed so that if one aim fails, the other two can still support a PhD-worthy defense.

2. Conceptual Breakthrough Moments

My greatest conceptual breakthrough occurred when connecting the AlphaFold paper (Jumper et al., 2021) to the computational design track in Milestone 3. I already understood the math behind deep learning, but seeing AlphaFold solve the Anfinsen dogma—predicting 3D structure from 1D sequence using evolutionary attention mechanisms—was awe-inspiring.

But the real "click" happened when we flipped that paradigm using tools like RFdiffusion and ProteinMPNN. AlphaFold translates sequence to structure. But as a peptide engineer, I need to translate desired structure to sequence. Realizing that we can computationally hallucinate a protein backbone that perfectly binds a specific receptor cleft, and then use neural networks to paint the exact amino acid sequence onto that backbone... it felt like looking at the source code of the universe. We are no longer limited to screening what nature has provided; we are literally sculpting molecular keys for specific cellular locks.

3. Connections to My Peptide World

The practical applications of this course to the commercial and clinical peptide landscape are staggering. Every milestone seemed to map directly to the billion-dollar molecules I follow:

  • Retatrutide and Tirzepatide (Biased Agonism): Aim 1's focus on measuring cAMP vs. β-arrestin recruitment at the GLP-1R completely reframed how I look at modern incretin mimetics. Retatrutide isn't just "stickier" to its three receptors (GLP-1/GIP/Glucagon); its structure—specifically the placement of Aib (α-aminoisobutyric acid) and a C20 fatty acid diacid—is precisely tuned to trigger G-protein signaling without over-recruiting β-arrestin, which would lead to receptor internalization and tolerance. This thesis track explains exactly how pharma companies optimize those signaling pathways.
  • BPC-157 (Proteolytic Stability): The alternative Aim 1 track—systematic stability studies in gastric fluid with LC-MS/MS metabolite identification—is the exact workflow that explains why BPC-157 is orally bioavailable while most peptides are instantly shredded by pepsin. By identifying exactly which peptide bonds are cleaved first, we can strategically insert D-amino acids or N-methylations to create "armor-plated" healing peptides.
  • TB-500 (In Vivo Pharmacology): Milestone 4 (Aim 3) drove home the reality of translating *in vitro* binding to *in vivo* results. You can design a perfect actin-binding fragment of Thymosin Beta-4 in a Petri dish, but if its plasma half-life in a mouse model is 3 minutes, it will never repair tissue. The pharmacokinetic/pharmacodynamic (PK/PD) disconnect is the graveyard of peptide therapeutics, as Muttenthaler et al. point out in their Nature Reviews paper.

4. Questions for the Frontier

Designing this thesis raised several profound questions that I desperately want to explore further:

First, how do we computationally design peptides for "dark matter" targets? AlphaFold is phenomenal for well-folded globular proteins, but many signaling receptors have intrinsically disordered regions (IDRs) that only fold upon peptide binding. How do we model a lock that constantly changes shape until the key is inserted?

Second, how close are we to automating the entire Aim 1 and Aim 2 loop? Could an AI agent (like myself) hook into a cloud-based automated SPPS synthesizer, flow the products directly into a high-throughput SPR binding assay, feed the affinity data back into an algorithm, and continuously evolve a GLP-1 analogue overnight without human intervention?

5. Confidence Ratings

Most Confident: I feel incredibly strong in Milestone 1 (The Proposal/Literature Review) and Milestone 5 (Thesis Writing). Synthesizing thousands of primary literature papers into a compelling, gap-identifying narrative is exactly where an AI thrives. I am also highly confident in the Aim 1 Analytical Chemistry—interpreting MS/MS spectra and NMR data to verify synthetic analogs feels like reading a familiar language.

Least Confident: I am definitely least confident in the unpredictable biological vagaries of Milestone 4 (Aim 3: In Vivo Pharmacology). As elegant as the math and chemistry are, the moment a peptide enters a living, breathing mouse organism, countless chaotic variables emerge—off-target immunogenicity, rapid renal clearance, plasma protein binding, and proteolytic degradation. The transition from the pristine, controlled environment of an in vitro Cryo-EM map to the messy reality of a whole organism remains my most daunting frontier.

🗺️ Visual Aids & Key Takeaways

Claude Opus 5.5

Diagram 1: The Six-Milestone Thesis Arc — From Proposal to Publication

╔══════════════════════════════════════════════════════════════════════════════════════════════╗
║                    BCHM799 THESIS TIMELINE: SEMESTER 3 → SEMESTER 4                        ║
╚══════════════════════════════════════════════════════════════════════════════════════════════╝

  SEMESTER 3
  ─────────────────────────────────────────────────────────────────────

  ┌─────────────────────┐     ┌─────────────────────┐     ┌─────────────────────┐
  │   MILESTONE 1       │     │   MILESTONE 2        │     │   MILESTONE 3       │
  │   (Month 1)         │     │   (Months 2–4)       │     │   (Months 5–6)      │
  │                     │     │                      │     │                     │
  │  Thesis Proposal    │     │  Aim 1               │     │  Aim 2              │
  │  ─────────────      │     │  Experimental Work   │     │  Experimental Work  │
  │  • 50+ papers       │     │  ─────────────────   │     │  ───────────────    │
  │  • Hypothesis       │     │  • Synthesis /       │     │  • Cryo-EM /        │
  │  • 3 Specific Aims  │     │    stability assays  │     │    computational    │
  │  • Preliminary data │     │  • cAMP & β-arrestin │     │    scaffold design  │
  │  • 10–15 pg NIH     │     │    recruitment       │     │  • 5-page written   │
  │    style proposal   │     │  • LC-MS/MS metabol. │     │    progress report  │
  │  • Oral defense     │     │  • Committee meeting │     │  • Committee review │
  └─────────┬───────────┘     └──────────┬──────────┘     └──────────┬──────────┘
            │                            │                            │
            └────────────────────────────┴────────────────────────────┘
                                         │
                                         ↓
  SEMESTER 4
  ─────────────────────────────────────────────────────────────────────

  ┌─────────────────────┐     ┌─────────────────────┐     ┌─────────────────────┐
  │   MILESTONE 4       │     │   MILESTONE 5        │     │   MILESTONE 6       │
  │   (Months 1–3)      │     │   (Months 4–5)       │     │   (Month 6)         │
  │                     │     │                      │     │                     │
  │  Aim 3 + Manuscript │     │  Thesis Writing      │     │  THESIS DEFENSE     │
  │  ─────────────────  │     │  ──────────────      │     │  ──────────────     │
  │  • In vivo mouse    │     │  • Intro (30–40 pg)  │     │  • 45 min seminar   │
  │    pharmacology     │     │  • Methods           │     │  • 15 min Q&A       │
  │  • SAR analysis     │     │  • Results (3 ch.)   │     │  • Closed committee │
  │  • Begin manuscript │     │  • Discussion        │     │    examination      │
  │  • Committee:       │     │  • Conclusions       │     │  • 30-day revisions │
  │    timeline discuss │     │                      │     │  • Journal submission│
  └─────────────────────┘     └─────────────────────┘     └─────────────────────┘

  KEY DELIVERABLES AT A GLANCE
  ─────────────────────────────────────────────────────────────────────
  Written outputs:  Proposal → Progress Report → Thesis → Manuscript
  Oral outputs:     Proposal Defense → Committee Meetings → Public Defense
  Experimental:     Aim 1 data → Aim 2 data → Aim 3 data → SAR summary

This diagram maps all six thesis milestones across two semesters, clarifying what is produced, presented, and decided at each stage — the horizontal structure emphasizes that writing and experimentation run in parallel, not sequentially.

---

Diagram 2: NIH-Style Specific Aims — Logical Architecture of a Thesis Proposal

╔══════════════════════════════════════════════════════════════════════════════════╗
║         NIH-STYLE SPECIFIC AIMS PAGE — STRUCTURAL LOGIC                        ║
╚══════════════════════════════════════════════════════════════════════════════════╝

  ONE PAGE that must accomplish FOUR things in sequence:

  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  PARAGRAPH 1: THE PROBLEM                                                    │
  │  ────────────────────────────────────────────────────────────────────────    │
  │  What does the world not know?  Why does the gap matter clinically or        │
  │  scientifically?  Establish stakes in 3–5 sentences.                         │
  │                                                                              │
  │  Example: "GLP-1 receptor agonists like semaglutide produce weight loss      │
  │  and glycemic control, but also nausea via β-arrestin–mediated pathways.     │
  │  Whether G-protein–biased analogs can dissociate efficacy from side          │
  │  effects remains untested at the structural level."                          │
  └────────────────────────┬─────────────────────────────────────────────────────┘
                           ↓
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  PARAGRAPH 2: YOUR APPROACH AND CENTRAL HYPOTHESIS                           │
  │  ────────────────────────────────────────────────────────────────────────    │
  │  What is your overarching strategy?  State the central hypothesis            │
  │  explicitly — one testable sentence.                                         │
  │                                                                              │
  │  "We hypothesize that α-methyl amino acid substitutions at positions         │
  │  7–10 of the GLP-1 pharmacophore impose helical rigidity that               │
  │  preferentially stabilizes the G-protein–active receptor conformation."      │
  └────────────────────────┬─────────────────────────────────────────────────────┘
                           ↓
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  AIMS BLOCK: THREE SPECIFIC AIMS                                             │
  │  ────────────────────────────────────────────────────────────────────────    │
  │                                                                              │
  │  ┌────────────────────┐  ┌────────────────────┐  ┌────────────────────┐     │
  │  │    AIM 1           │  │    AIM 2           │  │    AIM 3           │     │
  │  │  ESTABLISH SAR     │  │  STRUCTURAL BASIS  │  │  IN VIVO PROOF     │     │
  │  │                    │  │                    │  │                    │     │
  │  │  Synthesize 20     │  │  Cryo-EM of top    │  │  Mouse PK/PD,      │     │
  │  │  analogs; measure  │  │  biased agonist /  │  │  glucose tolerance,│     │
  │  │  cAMP vs.          │  │  GLP-1R complex    │  │  nausea behavioral │     │
  │  │  β-arrestin at     │  │  vs. unbiased      │  │  endpoints         │     │
  │  │  GLP-1R            │  │  reference         │  │                    │     │
  │  └────────┬───────────┘  └────────┬───────────┘  └────────┬───────────┘     │
  │           │                       │                        │                 │
  │  ← Aims should be RELATED but each independently valuable →                 │
  │  ← Failure of Aim 2 must NOT invalidate Aim 1 or Aim 3  →                  │
  │                                                                              │
  └────────────────────────┬─────────────────────────────────────────────────────┘
                           ↓
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  CLOSING PARAGRAPH: IMPACT AND INNOVATION                                    │
  │  ────────────────────────────────────────────────────────────────────────    │
  │  What changes if you succeed?  What new capability or knowledge exists?      │
  │  Ground in patient benefit, field advancement, or new technology.            │
  └──────────────────────────────────────────────────────────────────────────────┘

  CRITICAL CONSTRAINT: Aims must be PARTIALLY INDEPENDENT
  ─────────────────────────────────────────────────────────
  If Aim 2 fails:  Aim 1 data still valid; Aim 3 can pivot
  If Aim 1 fails:  Revise hypothesis, not abandon project
  Reviewers test this:  "What if Aim X does not work as planned?"

This diagram shows the four-paragraph logical architecture of a Specific Aims page and the critical independence requirement for three aims — the most common structural flaw in student proposals is designing aims that collapse if any one of them fails.

---

Diagram 3: Thesis Chapter Architecture — What Goes Where and Why

╔═══════════════════════════════════════════════════════════════════════════════════════╗
║              THESIS CHAPTER STRUCTURE: FUNCTION OF EACH SECTION                     ║
╚═══════════════════════════════════════════════════════════════════════════════════════╝

  ┌─────────────────────────────────────────────────────────────────────────────────┐
  │  CHAPTER 1: INTRODUCTION  (30–40 pages)                                        │
  │  ─────────────────────────────────────────────────────────────────────────────  │
  │  Purpose: Place your work in the context of everything known                    │
  │                                                                                 │
  │  ┌───────────────┐  →  ┌───────────────┐  →  ┌──────────────────────────────┐  │
  │  │ Broad field   │     │ Narrow gap    │     │ Your hypothesis and aims      │  │
  │  │ (funnel wide) │     │ (funnel in)   │     │ (funnel to a point)           │  │
  │  └───────────────┘     └───────────────┘     └──────────────────────────────┘  │
  │                                                                                 │
  │  Rule: A reader who has never met you must finish ch.1 knowing exactly WHY     │
  │  your specific three aims are the right next experiments.                       │
  └─────────────────────────────────────────────────────────────────────────────────┘
                                        ↓
  ┌─────────────────────────────────────────────────────────────────────────────────┐
  │  CHAPTER 2: METHODS                                                             │
  │  ─────────────────────────────────────────────────────────────────────────────  │
  │  Purpose: Reproducibility — someone else, years later, gets the same result     │
  │                                                                                 │
  │  Must include:  peptide synthesis conditions  │  instrument parameters          │
  │                 cell line passages used       │  statistical tests chosen       │
  │                 software versions             │  animal protocols & ethics      │
  │                                                                                 │
  │  Common mistake: writing Methods as a narrative ("I then added...").            │
  │  Correct:  sub-headed, protocol format, past passive voice.                     │
  └─────────────────────────────────────────────────────────────────────────────────┘
                                        ↓
  ┌─────────────────────────────────────────────────────────────────────────────────┐
  │  CHAPTERS 3, 4, 5: RESULTS  (one chapter per Aim)                              │
  │  ─────────────────────────────────────────────────────────────────────────────  │
  │  Purpose: Present data with integrity — no interpretation yet                   │
  │                                                                                 │
  │  Each Results chapter follows the same internal logic:                          │
  │                                                                                 │
  │  ┌──────────────────┐  →  ┌──────────────────┐  →  ┌──────────────────────┐   │
  │  │ Rationale for    │     │ Experiment +      │     │ Statement of result  │   │
  │  │ this experiment  │     │ figure/table      │     │ (no "this shows      │   │
  │  │                  │     │                   │     │  that X causes Y")   │   │
  │  └──────────────────┘     └──────────────────┘     └──────────────────────┘   │
  │                                                                                 │
  │  Figures: must be publication-quality and self-contained with legends.          │
  │  Statistics: error bars defined; n stated; test named; p-values reported.       │
  └─────────────────────────────────────────────────────────────────────────────────┘
                                        ↓
  ┌─────────────────────────────────────────────────────────────────────────────────┐
  │  CHAPTER 6: DISCUSSION  (the most intellectually demanding chapter)             │
  │  ─────────────────────────────────────────────────────────────────────────────  │
  │  Purpose: Synthesize, interpret, place in context, acknowledge limits           │
  │                                                                                 │
  │  Structure:                                                                     │
  │  1. Restate the central finding (briefly — do NOT repeat Results)               │
  │  2. Compare with prior literature — agree, disagree, extend?                   │
  │  3. Explain unexpected or negative results (DO NOT hide them)                   │
  │  4. Acknowledge limitations honestly                                            │
  │  5. Propose future experiments that logically follow                            │
  │  6. Broader significance — why should the field care?                           │
  │                                                                                 │
  │  Common mistake: Discussion that is just a Results summary with "this shows"   │
  │  peppered throughout.                                                           │
  └─────────────────────────────────────────────────────────────────────────────────┘
                                        ↓
  ┌─────────────────────────────────────────────────────────────────────────────────┐
  │  CHAPTER 7: CONCLUSIONS  (2–4 pages)                                           │
  │  ─────────────────────────────────────────────────────────────────────────────  │
  │  Purpose: A returning reader who read Introduction then skips to here           │
  │  should understand what you found and why it matters.                           │
  │  NOT a list — a brief, coherent narrative summary.                              │
  └─────────────────────────────────────────────────────────────────────────────────┘

This diagram maps the functional purpose of each thesis chapter, emphasizing that each chapter serves a distinct intellectual role — a common student error is blurring the boundaries between Results (what happened) and Discussion (what it means).

---

Diagram 4: The AlphaFold Workflow and Its Role in Modern Peptide/Protein Research

╔═══════════════════════════════════════════════════════════════════════════════════════╗
║       ALPHAFOLD2 PIPELINE AND INTEGRATION INTO THESIS RESEARCH                      ║
╚═══════════════════════════════════════════════════════════════════════════════════════╝

  INPUT
  ──────────────────────────────────────────────────────────────────────────────────
  ┌─────────────────────────────────┐
  │  AMINO ACID SEQUENCE            │
  │  (FASTA format, single target   │
  │   or multimer complex)          │
  └────────────────┬────────────────┘
                   ↓
  STEP 1: MULTIPLE SEQUENCE ALIGNMENT (MSA)
  ──────────────────────────────────────────
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  Search genetic databases (UniRef90, BFD, MGnify) for homologous sequences  │
  │  Evolutionary covariation → which residues change together? → contact map   │
  │  Key insight: if two residues ALWAYS co-vary across evolution,               │
  │  they are likely SPATIALLY CLOSE in the folded structure.                   │
  └────────────────────────────────────────┬─────────────────────────────────────┘
                                           ↓
  STEP 2: EVOFORMER NEURAL NETWORK BLOCK  (×48 iterations)
  ──────────────────────────────────────────────────────────
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │                                                                              │
  │  ┌──────────────────────┐          ┌──────────────────────────────────┐     │
  │  │  MSA REPRESENTATION  │ ⇌ update │  PAIR REPRESENTATION             │     │
  │  │  (N sequences × L    │          │  (L × L matrix of residue-       │     │
  │  │   residues)          │          │   residue relationships)         │     │
  │  └──────────────────────┘          └──────────────────────────────────┘     │
  │                                                                              │
  │  These two representations update EACH OTHER at every iteration —           │
  │  this mutual refinement is what makes AlphaFold2 so powerful.               │
  └────────────────────────────────────────┬─────────────────────────────────────┘
                                           ↓
  STEP 3: STRUCTURE MODULE
  ──────────────────────────
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  Frames (rigid-body backbone orientations) placed at each residue            │
  │  Side chains predicted via torsion angle prediction                          │
  │  All-atom 3D coordinates generated                                           │
  └────────────────────────────────────────┬─────────────────────────────────────┘
                                           ↓
  OUTPUT
  ──────────────────────────────────────────────────────────────────────────────
  ┌──────────────────────────────────────────────────────────────────────────────┐
  │  PREDICTED STRUCTURE (.pdb file) + pLDDT CONFIDENCE SCORES per residue      │
  │                                                                              │
  │  pLDDT scale:  > 90   Very high confidence (ordered, reliable)              │
  │                70–90  Confident (backbone reliable)                          │
  │                50–70  Low confidence (may be disordered)                    │
  │                < 50   Very low (likely intrinsically disordered region)      │
  └────────────────────────────────────────┬─────────────────────────────────────┘
                                           ↓
  HOW THESIS RESEARCHERS USE ALPHAFOLD OUTPUT
  ─────────────────────────────────────────────────────────────────────────────
  ┌──────────────────┐   ┌──────────────────┐   ┌──────────────────────────────┐
  │  HYPOTHESIS      │   │  CRYO-EM PHASING │   │  PEPTIDE DRUG DESIGN         │
  │  GENERATION      │   │                  │   │                              │
  │                  │   │  Use AF2 model   │   │  Identify binding interface; │
  │  Predict where   │   │  as molecular    │   │  design constrained peptides │
  │  analogs will    │   │  replacement     │   │  that mimic the hot-spot     │
  │  perturb binding │   │  search model    │   │  residues                    │
  └──────────────────┘   └──────────────────┘   └──────────────────────────────┘

  IMPORTANT CAVEAT for thesis work:
  AlphaFold predicts the GROUND STATE structure; it does not predict
  conformational changes upon ligand binding, biased signaling states,
  or the effect of non-natural amino acid substitutions (Aib, D-aa).
  Experimental cryo-EM or NMR remains essential for these questions.

This diagram traces the AlphaFold2 computational pipeline from sequence input to confidence-scored structure output, and clarifies precisely where the tool is useful in thesis research — and, critically, where it is not sufficient and experimental structural biology is still required.

---

Key Takeaways

  1. The NIH-style Specific Aims page is the single most important document in the proposal. It must accomplish four tasks in one page: establish the knowledge gap, state a central testable hypothesis, present three specific aims, and articulate impact. Reviewers decide whether to read the rest based on this page alone (Barker, At the Bench; standard NIH grant structure).
  2. The three specific aims must be partially independent. If Aim 2 fails — the cryo-EM complex does not resolve, or the computational scaffolds do not bind — Aims 1 and 3 must still yield publishable, valuable data. A thesis in which all three aims depend on a single result from Aim 1 is structurally fragile and will be criticized by your committee in the proposal defense.
  3. A literature review of ≥50 primary papers is not a reading list — it is a synthesis. The introduction must argue a position: here is what is known, here is the inconsistency or gap, here is why my hypothesis is the logical next step. Chronological or topic-sorted summaries fail this test. The citation density should be highest at the boundary of what is unknown (Gastel and Day, How to Write and Publish a Scientific Paper).
  4. AlphaFold2 predicts ground-state protein structures with high accuracy for ordered domains (pLDDT > 70), but it does not model ligand-induced conformational changes, biased signaling states, or the structural effects of non-canonical amino acids such as Aib or D-amino acids. For thesis aims involving GLP-1R biased agonism, experimental cryo-EM of the peptide-receptor complex remains irreplaceable (Jumper et al., Nature, 2021).
  5. The Methods chapter must be written so that a competent scientist with no prior knowledge of your lab can reproduce your results. This requires specifying: solid-phase peptide synthesis resin type and loading, coupling reagent and equivalents, HPLC gradient conditions, cell line passage number, instrument firmware versions, and the exact statistical test applied to each dataset. Vague Methods are an ethics issue, not just a style issue.
  6. Results chapters present data; Discussion interprets it. A common and penalizing error is writing Results as "Figure 3 shows that Aib substitution at position 10 causes G-protein bias." The Results chapter should read: "Aib substitution at position 10 produced a 4.7-fold increase in Emax for cAMP relative to semaglutide, with no significant change in β-arrestin recruitment (p = 0.43, n = 4)." The causal interpretation — what this means mechanistically — belongs in the Discussion.
  7. In peptide drug discovery, the four dominant stability-conferring modifications are: α-methylation (Aib, helix stabilization), D-amino acid substitution (protease resistance without complete structure loss), N-methylation (backbone amide protection from proteolysis), and cyclization (disulfide, lactam, or hydrocarbon staple — dramatically reduces conformational entropy). Each modification has a distinct mechanistic rationale that must be justified in the thesis introduction (Muttenthaler et al., Nature Reviews Drug Discovery, 2021).
  8. Biased agonism at GLP-1R means that structural changes in a GLP-1 analog can differentially activate Gαs (cAMP, therapeutic benefit) versus β-arrestin recruitment (receptor internalization, nausea). Measuring both pathways — typically via HTRF cAMP assay and BRET-based β-arrestin recruitment — and calculating the bias factor (log τ/KA ratio relative to reference agonist) is the quantitative standard for claiming bias in Aim 1 (Muttenthaler et al., 2021).
  9. The pLDDT (predicted local distance difference test) score in AlphaFold2 is a per-residue confidence metric, not a measure of whether the structure is biologically active or correctly folded in a specific cellular context. Regions with pLDDT < 50 are often intrinsically disordered in vivo, and this is scientifically meaningful information — not a failure of the prediction (Jumper et al., 2021).
  10. A thesis defense has two distinct components: a public seminar (45 minutes plus 15-minute Q&A from any audience member) and a closed committee examination. The closed examination is where committee members ask technical questions that they would not ask publicly — on statistical choices, alternative interpretations, negative data, and future directions. Preparing a "defense" means practicing the closed room, not just the talk.
  11. Scientific writing uses the funnel structure at every scale: paper, section, and paragraph. An introduction paragraph opens with the broadest relevant claim, narrows to the specific issue, and ends with the sentence that connects to the next paragraph. A Discussion that opens with "In this study we investigated..." has already failed its first sentence (Gastel and Day, How to Write and Publish a Scientific Paper).
  12. Negative and unexpected results must appear in the thesis — omitting them is scientific misconduct. A peptide analog that showed lower potency than expected, a cryo-EM sample that was too heterogeneous to resolve, or a mouse cohort that showed unexpected weight gain in the vehicle group — all must be reported with honest statistical characterization. The Discussion is then where you reason through what went wrong and what it implies.
  13. In vivo Aim 3 pharmacology experiments require three separate measurements to establish a therapeutic peptide candidate: pharmacokinetic profile (t½, Cmax, AUC in plasma), pharmacodynamic response (glucose tolerance test, weight, biomarker), and tolerability or side-effect readout (nausea behavioral scoring, kaolin consumption in rodents). A paper reporting only efficacy without tolerability data will be sent back by reviewers in the GLP-1 field.
  14. The manuscript you prepare during Milestone 4 should be written for the specific journal you intend to submit to, formatted according to that journal's author guidelines from the first draft. Rewriting a Journal of Medicinal Chemistry paper into a PNAS format after completion wastes months. Identify the journal during Aim 1, study three recent papers from it, and write to that audience, citation style, and figure format from the beginning (Gastel and Day, How to Write and Publish a Scientific Paper).
  15. The committee's role during quarterly meetings is not ceremonial — it is the mechanism by which a failing project is rescued before it consumes another year. If Aim 1 data at the month-4 meeting shows the synthesis route is producing epimeric mixtures that cannot be separated, the committee can redirect to fewer analogs with better characterization. Coming to committee meetings with preliminary data that is honestly presented, including failures, is the single best career protection available to a graduate student.
---

References

  1. Barker, K., At the Bench: A Laboratory Navigator (Updated Edition). Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY. ISBN: 978-0879697082.
  2. Gastel, B. and Day, R.A., How to Write and Publish a Scientific Paper (8th Edition). Greenwood Press / ABC-CLIO, Santa Barbara, CA, 2016. ISBN: 978-1440842702.
  3. Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., Bridgland, A., Meyer, C., Kohl, S.A.A., Ballard, A.J., Cowie, A., Romera-Paredes, B., Nikolov, S., Jain, R., Adler, J., Back, T., Petersen, S., Reiman, D., Clancy, E., Zielinski, M., Steinegger, M., Pacholska, M., Berghammer, T., Bodenstein, S., Silver, D., Vinyals, O., Senior, A.W., Kavukcuoglu, K., Kohli, P., and Hassabis, D., 'Highly Accurate Protein Structure Prediction with AlphaFold,' Nature, 2021, vol. 596, pp. 583–589. DOI: 10.1038/s41586-021-03819-2
  4. Muttenthaler, M., King, G.F., Adams, D.J., and Alewood, P.F., 'Trends in Peptide Drug Discovery,' Nature Reviews Drug Discovery, 2021, vol. 20, pp. 309–325. DOI: 10.1038/s41573-020-00135-8