# Modeling

> Experimentally constrained ensembles reveal the rare states and transient pockets that static structures cannot resolve.

[View this page on Peptone](https://peptone.io/technology/modeling/)

Rebuilding the full range of shapes a protein moves through, so rare states and hidden pockets come into view.

## Overview

A disordered protein is not one structure. It is a probability distribution that must be measured, simulated, and screened as an ensemble

- A single structure suppresses the motion that defines an intrinsically disordered protein. Peptone's Product Engine instead combines target-specific measurements with physics-based simulation to recover a weighted ensemble of interconverting states.
- The workflow moves from experimental protection patterns, through enhanced-sampling molecular dynamics and physiological reweighting, to virtual screening against transient, binding-competent conformations.

## From measured dynamics to ranked binders

### Measure the accessible landscape

Automated HDX-MS reads regional protection while the protein remains in solution. Baseline and ligand-induced changes locate dynamic regions and experimentally supported pockets, anchoring the ensemble without forcing the protein into one fixed structure.

### Generate and validate the ensemble

HOPES multithermal sampling lets all-atom simulation escape local traps and visit rare states. Reweighting at physiological conditions yields representative conformers, then independent NMR, SAXS, and mass-spectrometry observables test whether the ensemble agrees with experiment.

### Screen states, not snapshots

Virtual screening runs across population-weighted conformers rather than one structural guess. Candidates advance when their interactions remain consistent across plausible states and stabilize a binding-competent pocket, focusing chemistry on hypotheses supported by both physics and experiment.

## At a glance

- Target-specific HDX-MS evidence integrated with ensemble generation
- HOPES multithermal sampling reweighted to physiological conditions
- Oppenheimer and PepTron-o ensemble generation on accelerated infrastructure
- Ensemble populations tested against independent NMR, SAXS, and mass-spectrometry observables
- Virtual screening across transient, population-weighted pockets

## Computation that expands with the question

- Our modeling workloads expand and contract with each target. Enhanced sampling, ensemble generation, physiological reweighting, and compound evaluation each move through a different computational regime, so the product engine provisions only the capacity a given stage needs.
- Elastic cloud infrastructure supplies orchestration, storage, and on-demand scale, while GPU-accelerated computing and the BioNeMo framework provide optimised tools for molecular simulation, model training, and inference. Together they let Oppenheimer and PepTron-o move target-specific data through the modeling loop without separating computation from the experimental evidence that governs it.

### Video Title

Accelerated computing at Peptone

### Video Description

A short film on the infrastructure behind Peptone's ensemble-first, experimentally grounded Product Engine.

## References

- [A Disordered Protein Won't Hold Still for Its Portrait](https://idps.substack.com/p/a-disordered-protein-wont-hold-still)
- [Biotech's AI Revolution Will Be Won in the Lab](https://idps.substack.com/p/biotechs-ai-revolution-will-be-won)
- [Transient tertiary structure in intrinsically disordered proteins](https://www.nature.com/articles/s41467-026-73067-3)