A third of the human proteome never folds into a fixed structure. Its biology, and much of human disease, lives in that motion. Peptone is built to read it, model it, and drug it.
Most medicines are designed against proteins that hold a fixed shape. Much of what drives human disease never does.
For a century, biology has read a straight line from sequence to structure to function: a chain of amino acids folds into one shape, and that shape does the work. It holds for a large part of the proteome and breaks for the rest. Roughly a third of the residues in the human proteome never fold into a stable structure, and most human proteins carry at least one disordered region. These intrinsically disordered proteins are not broken or unfinished. They do their job precisely because they stay in motion.
Prevalence of intrinsic disorder in the human proteome. About 33% of all residues are predicted never to fold into a stable structure, and close to 60% of human proteins carry at least one disordered region of 30 residues or more. Estimates from PritiĊĦanac et al., A Functional Map of the Human Intrinsically Disordered Proteome (2024).
A folded protein settles into a single cooperative structure and presents a stable pocket, the kind of feature a drug can be designed to fit. An intrinsically disordered protein has a defined sequence but no single folded form. It moves continuously through a broad ensemble of interconverting conformations, and that flexibility is the mechanism rather than a defect in it. For these proteins the ensemble is the molecule. A single snapshot does not describe them, and can be actively misleading.
Disordered regions concentrate at the busiest junctions of the cell: the transcription factors that decide which genes switch on, the signaling proteins that relay messages, and the membraneless condensates that organize the cell interior. Because these roles are central, their failure is severe. Trace many of the most stubborn diseases to their molecular origin and a disordered protein is often waiting there. That same biology defines where we work: oncology, CNS, immunology and inflammation, and longevity.
Structural biology was built around methods that need a static, cooperative subject. X-ray crystallography and cryo-electron microscopy resolve a well-ordered structure beautifully and a moving ensemble poorly, so disordered proteins were systematically excluded from the Protein Data Bank. Only a few techniques could observe them in solution at all, chief among them nuclear magnetic resonance, supported by small-angle X-ray scattering. With little to see and less to design against, most of the disordered proteome was set aside as undruggable.
Archive coverage is schematic. Structure counts and methods: RCSB Protein Data Bank.
The prediction models that reshaped structural biology, from AlphaFold2 and AlphaFold3 to open successors such as OpenFold, Chai-1, and Boltz, all learned from that same archive. Given a sequence they return a single confident, static structure. Ask one for a disordered target, such as the androgen receptor N-terminal domain, and it returns a crisp, convincing shape that the protein never actually holds. The models even flag the problem themselves: their per-residue confidence collapses across the disordered regions, marking the residues that matter most as the least reliable. More data of the same kind does not close the gap.
Based on the AlphaFold prediction (AF-P10275) of the androgen receptor; colouring follows the pLDDT model-confidence scale.
Because the data does not exist in public archives, we generate it. Peptone reads a target where it actually lives, in solution and in motion, using hydrogen-deuterium exchange mass spectrometry together with magnetic and paramagnetic resonance spectroscopy, including NMR, PRE, and EPR. These experiments are sparse by design and decisive in effect. A small number of well-chosen restraints steer simulation and generative models away from plausible fiction and toward physical reality, collapsing billions of candidate conformations down to the thermodynamically sensible states a protein truly populates.
The payoff is a state that no single structure would ever show. Applying multithermal enhanced sampling to ACTR, a well-characterized disordered protein, resolved a rare conformation populated only about three percent of the time: transiently folded, reached across a modest barrier of roughly 13 kilojoules per mole, and competent to bind. The computed ensemble agreed with independent NMR and SAXS measurements. A binding-competent pocket that appears in none of the protein's average structures becomes a defined, addressable target. This is how sparse experiment redirects frontier methods and drives them quickly to physically relevant answers.
Atomistic ensembles of ACTR were generated with multithermal enhanced sampling (On-the-fly Probability Enhanced Sampling, OPES), then reweighted against NMR chemical shifts and paramagnetic relaxation enhancements and independently cross-validated with residual dipolar couplings and small-angle X-ray scattering (SAXS). The reweighted ensemble resolves a rare, bound-like state (about 3%) in which the binding helices pack together, invisible to any single average structure. Streit, Invernizzi, Bottaro, Tamiola and Lindorff-Larsen, Nature Communications 17, 5558 (2026).
None of this is method for its own sake. Reading real biophysics instead of confident artifacts tightens every design, measure, and decide cycle, and each turn of the loop sharpens the next. The purpose is singular: to generate drugs that stand the scrutiny of clinical trials, against targets that structure-first discovery leaves untouched.