How New Peptides Are Discovered

Three ways to find a new peptide

Every peptide that has ever entered a lab notebook came from one of three broad strategies, or some combination of them. The first is isolation from nature: purifying a peptide that a living system already makes, such as a hormone from an animal gland or a toxin from venom. The second is rational design: starting from a known target or a natural ligand and deliberately engineering a new molecule to bind it. The third is high-throughput selection: building an enormous library of candidate sequences and letting a binding assay pick out the rare winners. Modern discovery rarely uses just one of these — it braids them together.

What unites all three routes is a shared dependence on chemistry and, increasingly, computation. You cannot pursue a designed peptide without a way to make it, and you cannot make sense of a billion-member library without tools to read out which sequences succeeded. This is why solid-phase peptide synthesis and modern sequencing sit at the center of the story, and why AI-driven design has moved so quickly from novelty to standard practice. If you are new to the molecules themselves, our primer on what peptides are is a useful starting point.

It also helps to be precise about what "discovery" means. Finding a peptide that binds a target is only the opening move. The long middle game — turning a fragile, low-potency hit into a stable, selective, manufacturable candidate — is where most of the real work happens. This guide follows that whole arc, from the first source of a sequence to the point where a candidate is ready to be evaluated as a potential therapeutic.

Source one: isolating peptides from biology

The oldest and still one of the most important sources of new peptides is nature itself. The body is full of signaling peptides — hormones, neuropeptides, and growth factors — and for most of the twentieth century, discovering them meant physically purifying them from tissue. This was heroic, laborious chemistry. Researchers processed vast quantities of animal glands to extract vanishingly small amounts of the active molecule, then worked out its sequence and structure.

The hypothalamic-releasing hormones are the classic example. Isolating factors such as thyrotropin-releasing hormone and gonadotropin-releasing hormone required processing enormous numbers of animal hypothalami to obtain milligram quantities of pure peptide — work that ultimately earned Nobel recognition. The same tissue-extraction spirit gave the world insulin in the 1920s and, later, defined the endocrine peptides that anchor whole therapeutic areas today. Our overview of the history of peptide research traces this era in detail.

Isolation is not just a historical footnote. Even now, characterizing an endogenous human peptide can reveal a validated drug target — a receptor the body already uses — and a natural ligand to engineer from. Many of the most successful modern peptide drugs are refined descendants of a natural hormone: change a few residues to resist degradation, and a fleeting signal becomes a durable medicine. That logic connects directly to the incretin story explored in our guide to multi-agonist peptides.

Source two: venoms and natural products

If endogenous hormones are nature's signaling toolkit, venoms and other natural products are nature's combinatorial chemistry set. Venomous animals — snakes, cone snails, spiders, scorpions, and others — have evolved complex mixtures of peptides that act with extraordinary precision on ion channels, receptors, and enzymes. Evolution has effectively run a multimillion-year optimization campaign, and the survivors are potent, selective molecules that would be very hard to design from first principles.

Cone snail venoms are a celebrated case. These marine snails paralyze prey with cocktails of small, highly structured peptides known as conotoxins, many of which target specific ion channels with remarkable selectivity. That selectivity is exactly what a drug hunter wants, because it can translate into a strong effect with fewer off-target consequences. Natural-product peptides from bacteria and fungi have likewise seeded whole classes of antimicrobial and immunosuppressant chemistry.

The immune system provides another rich vein of naturally occurring peptides. Antimicrobial peptides such as LL-37, part of human innate immunity, disrupt microbial membranes through mechanisms distinct from conventional antibiotics — which is why they draw so much interest against drug-resistant organisms. You can read more in our LL-37 research profile. Natural AMPs also give computational models a deep library of real sequences to learn from, tying this source directly back into modern AI design.

Source three: rational and structure-based design

Where isolation reads sequences out of biology, rational design writes them deliberately. The approach starts from knowledge — the structure of a target protein, the sequence of a natural ligand, or the shape of a binding pocket — and reasons toward a molecule that should engage it. In its simplest form, rational design means taking a natural peptide and systematically changing residues to improve a property: swap an easily degraded amino acid for a protease-resistant one, add a residue that anchors the peptide to a stabilizing partner, or trim the sequence to its active core.

Structure-based design goes further by using three-dimensional information about the target. If you know the shape of a receptor's binding site, you can design a peptide whose backbone and side chains complement it — the molecular equivalent of cutting a key to fit a lock. This is where structure-prediction tools have been transformative: reliable models of a target turn design from guesswork into an engineering problem, a shift covered in depth in our article on AI peptide discovery.

Rational design is also the engine behind some of the field's biggest successes. Engineering a native incretin hormone for a longer half-life, or combining the activities of two hormones into a single molecule, are both design problems solved by careful, iterative modification. The most ambitious version — hitting several receptors with one purpose-built sequence — is the subject of our companion guide to the science of multi-agonist peptides.

Display technologies: phage and mRNA display

When you do not know which sequence will bind a target, the most powerful strategy is to test a huge number at once and let the target itself select the winners. Display technologies make this possible by solving a deceptively hard problem: after you find a peptide that binds, how do you know its sequence so you can make more of it? The answer is to physically link every peptide to the genetic information that encodes it.

Phage display

In phage display, a library of peptide sequences is genetically fused to a coat protein of a bacteriophage — a virus that infects bacteria. Each phage particle carries a peptide on its surface and the corresponding gene inside. Expose the library to an immobilized target, wash away everything that does not stick, and the phages that remain are the binders. Because each one carries its own gene, you can amplify them in bacteria and sequence them to read out the winning peptides. This technique was foundational enough to be recognized with a Nobel Prize in Chemistry.

mRNA display

mRNA display takes the same link-peptide-to-gene idea further and does it entirely in a test tube. Here each peptide is covalently attached to its own messenger RNA through a small linker, and the whole process happens without living cells. Because it is cell-free, mRNA display can access astronomically large libraries — far bigger than what a bacterial system will tolerate — and it tolerates non-natural amino acids and cyclization chemistries that expand the accessible chemical space. That combination makes it especially good at finding constrained, drug-like macrocyclic peptides.

DNA-encoded libraries and high-throughput selection

Display technologies rely on biology to link a molecule to its code. DNA-encoded libraries (DELs) apply the same principle to purely chemical synthesis. Each molecule in the library is tagged with a unique DNA "barcode" that records exactly how it was built — which chemical building blocks were added, and in what order. Because the tag is just DNA, a single tube can hold libraries numbering in the millions or billions of distinct compounds.

Screening a DEL is elegant: pour the whole mixed library over an immobilized target, wash away the non-binders, then use DNA sequencing to read the barcodes of whatever stuck. The barcodes tell you the chemical history of the winners, so you can re-synthesize the promising structures on their own for validation. Because DELs are built by chemistry rather than biology, they can include non-natural amino acids, unusual linkers, and cyclization strategies that broaden the reachable chemical space beyond what ribosomes produce.

All of these selection methods share a common philosophy: instead of guessing which molecule will work, build enormous diversity and let a binding assay do the choosing. The trade-off is that a raw hit from any library is exactly that — a hit, not a finished molecule. It may bind weakly, degrade quickly, or lack selectivity. Turning it into something useful is the job of the next stage.

The enabling chemistry: solid-phase peptide synthesis

None of the strategies above would be practical without a reliable way to make defined peptide sequences on demand. That capability arrived in 1963, when Bruce Merrifield introduced solid-phase peptide synthesis (SPPS) — an advance later recognized with the Nobel Prize in Chemistry. Before SPPS, making even a short peptide was a slow, low-yield ordeal. Merrifield's insight was to anchor the growing peptide chain to an insoluble solid support so that reagents and byproducts could simply be washed away between steps.

In SPPS, the peptide is built one amino acid at a time on a resin bead. Each cycle adds a protected amino acid, then removes the protecting group to expose the next attachment point, with washing steps in between to purify the intermediate. Repeating the cycle assembles the full sequence, which is finally cleaved from the resin. The process is so systematic that it was among the first chemistry to be automated, and modern synthesizers can assemble complex sequences far faster than any manual method.

SPPS is the quiet enabler of the entire discovery pipeline. It lets designers make the molecules they draw, lets teams re-synthesize hits pulled from libraries, and underpins the medicinal-chemistry cycles of optimization described below. It is also why peptides are such an attractive drug class in the first place: unlike large biologics grown in cells, a peptide of defined sequence can be manufactured chemically with tight control over purity and identity — the qualities documented on a certificate of analysis when research material is sourced.

From hit to lead: optimizing a raw discovery

A newly discovered peptide — whether isolated, designed, or selected — is almost never ready for prime time. Native peptides in particular are notoriously fragile: they are cleared from the body within minutes by peptidases and the kidneys, and they may bind their target too weakly or hit related receptors indiscriminately. Hit-to-lead optimization is the systematic process of fixing these shortcomings without breaking the activity that made the hit interesting.

• Potency and selectivity — residues are tuned to strengthen binding to the intended target while reducing activity at related receptors, improving the therapeutic window.
• Protease resistance — vulnerable amino acids are swapped for non-natural or D-amino acids, or the backbone is modified, so degrading enzymes no longer recognize the cleavage site.
• Half-life extension — strategies such as lipidation (attaching a fatty acid to promote albumin binding), PEGylation, or fusion are used to slow clearance, as detailed in our guide to peptide half-life.
• Constraint and cyclization — cyclizing a peptide locks it into its active shape, boosting both stability and binding affinity.
• Solubility and formulation — sequence tweaks reduce aggregation and improve handling, which also matters for stability and storage.

This is medicinal chemistry as an iterative loop: make a batch of analogs, test them, learn which changes helped, and design the next round. Each cycle is guided by structure-activity relationships — the accumulating map of which parts of the molecule matter for which property. Over many rounds, a weak, short-lived hit can become a potent, durable, selective lead. Increasingly, machine-learning models help prioritize which analogs to make, tightening the loop further.

From candidate to clinic

Optimization produces a lead, but a lead is still a long way from a medicine. Before a peptide can be tested in humans, it must clear extensive preclinical work: pharmacology to confirm the mechanism, pharmacokinetics to map how the body handles it, and toxicology to probe for safety liabilities. Manufacturing must be scaled up under strict quality controls so that every batch is consistent — the same identity and purity issues that a certificate of analysis is designed to document.

Only after that groundwork can a candidate enter clinical trials, which proceed in phases to evaluate safety, dosing, and effectiveness in progressively larger groups of people. This part of the journey is measured in years and has a high failure rate, and it is exactly the stage that discovery-stage acceleration does not shorten. Faster discovery gets you to a strong candidate sooner; it does not shortcut proving that the candidate is safe and effective.

That distinction is the crux of how this site frames the difference between research peptides and prescription peptides. A peptide can be genuinely exciting in a discovery paper and still be years — or forever — away from approval. Understanding the full pipeline is the best antidote to hype in either direction. For a sense of what is actually advancing beyond the bench, browse the cited profiles in our research library.