Chapter 35 — Further Reading

A note before the list. This chapter covers four discovery routes spanning roughly forty years, and the literature on each is enormous. The sources below are chosen to be readable and checkable — you should be able to reach most of them without an institutional subscription, and where you cannot, the abstract usually carries the point. Where a source is likely to be behind a paywall, it is flagged.

Everything here is background. None of it is guidance about taking anything, and several of the molecules discussed are approved drugs whose use is a matter for a clinician.

Tier 1 — Start here

Primary and near-primary accounts of the discoveries in this chapter. These are the sources that let you check the chapter rather than trust it.

• The original exendin-4 characterization. Search the literature for the early-1990s papers describing the isolation of an exendin peptide from Heloderma suspectum venom and its activity at the GLP-1 receptor. Read them for what they do not claim: the framing at the time is pharmacological curiosity, not drug development, and the gap between that framing and the eventual 2005 approval is instructive on its own.
• The regulatory documents for exenatide (2005). Approval packages are underrated reading. They contain the trial designs, the endpoints as pre-specified, the reviewers' objections, and the duration-of-action data that explains why liraglutide and semaglutide displaced it. Freely available from major regulators' public document systems.
• Reviews of the captopril discovery written by the people who did it. Ondetti and Cushman both wrote retrospective accounts of the reasoning from bradykinin-potentiating peptides through teprotide to captopril. These are among the clearest published descriptions of how a peptide lead becomes a small-molecule drug, and they are notably candid about how much was inference and how much was luck.
• The Nobel Prize lectures and popular summaries for Chemistry 2018 (phage display and directed evolution) and Chemistry 2024 (structure prediction and computational protein design). The Nobel Foundation publishes both a technical background document and a plain-language summary for each prize, free of charge. The technical backgrounds are the single most efficient way to get an accurate, non-promotional account of what was and was not achieved.
• The CASP assessment papers for CASP14 (2020). Written by the assessors, not the competitors. If you read only one item in this entire list, consider making it this one — it is the primary evidence behind the ✅ rating in §35.8, and reading an assessment written by people whose job is scoring rather than winning is a useful corrective to every press account of the same result.

Tier 2 — Go deeper

For readers who want the mechanisms, the methods, and the counterarguments.

• Reviews of venom peptides as drug leads. The toxinology literature has produced a steady stream of review articles cataloging venom-derived compounds in development. Read at least two from different decades — the shift in tone between older and newer reviews tells you something about how long "promising" can persist.
• Structural and functional reviews of conotoxins. These explain the disulfide-stapled architecture described in §35.2, the sheer diversity within a single Conus venom, and the channel-subtype selectivity that makes conotoxins such precise pharmacological tools.
• Ziconotide clinical literature and prescribing information. Read for the therapeutic window, the adverse effect profile, and the practicalities of intrathecal delivery. This is the most concrete illustration in medicine of what Chapter 4's blood-brain barrier constraint actually costs.
• Antimicrobial peptide reviews, ideally spanning the field's history. The magainin discovery and the subsequent decades of clinical difficulty are the best available case study in the gap between a mechanistically attractive idea and an approved medicine. Chapter 25 covers this; the reviews give you the detail.
• Methodological reviews of phage, ribosome, and mRNA display. Look for ones that discuss failure modes — plate binders, tag binders, propagation-advantage artifacts. A review that describes only successes is a marketing document.
• Cryo-EM structures of class B GPCRs. Many are deposited in public structural databases with accompanying papers. Even without structural biology training, the figures showing the peptide ligand threaded into the receptor's extracellular domain make §35.7's argument visually obvious in a way prose cannot.
• The ProteinMPNN and RFdiffusion papers, and independent replications of them. Read the original reports for what was achieved, then specifically seek out papers from other groups applying the methods. Independent reproduction is what moved this from "interesting preprint" to a defensible 🔬, and it is the thing press coverage never mentions.
• Analyses of clinical trial attrition and its causes. Several groups have published breakdowns of why compounds fail and at which phase. These are the empirical foundation of §35.10, and they are more sobering read in full than summarized. (Some are paywalled; the figures are usually reproduced in freely available commentary.)

Tier 3 — For the specialist and the skeptic

Harder, more technical, or more contested. Approach these when you want to argue with the chapter rather than absorb it.

• Critical commentary on AI in drug discovery, from both directions. There is now a real literature of people arguing that computational discovery will transform approval rates and people arguing that it cannot. Read at least one strong piece from each side, and note which side makes falsifiable predictions. That observation alone is worth the reading time.
• Analyses tracking the clinical progress of computationally derived candidates. These are the early data on the study specified in §35.8's evidence rating. They are currently too sparse to settle anything, which is precisely why the rating is ⚠️ rather than ✅ or ❌ — and watching a dataset accumulate toward a pre-specified threshold is excellent practice.
• Technical literature on intrinsically disordered proteins and peptides. This is the field that most directly underwrites §35.8's second limitation. It is genuinely difficult material and it will change how you read any claim about "the structure" of a short peptide.
• The structure-prediction limitations literature. Papers examining where predictions fail — point mutations, conformational change, complexes, disordered regions, orphan sequences. Written largely by people who use the tools daily and are not trying to debunk them, which makes them more useful than either the hype or the backlash.
• Toxinology methods literature on venom collection, fractionation, and screening. Rarely read outside the field, and it makes concrete how much labor sits behind the phrase "identified in the venom of."
• Historical accounts of the protein folding problem before 2020. Read one, and read it after everything else here. Nothing else conveys as well how large the change was — and understanding the size of a real achievement is a prerequisite for describing its limits without sounding like a cynic.

If you only do one thing

Read the CASP14 assessment written by the assessors, and then read a newspaper article about the same result.

Put them side by side. Both are describing something real; neither is lying. But the assessment states a scope — single domains, these targets, this metric — and the article states a transformation. Watch exactly where the scope drops out, and notice that it usually drops out in a single transitional sentence that nobody would flag as the error.

That is the same move you have been learning to catch for thirty-five chapters, performed on the most credentialed scientific result of the decade. If you can see it there, you can see it anywhere — and you will have learned it on a case where the underlying science is genuinely excellent, which is much harder, and much more useful, than learning it on a case where somebody was simply wrong.