# Design

> Multiple AI models explore sourceable chemistry against transient pockets in rare disordered-protein states, with matched HDX-MS closing the experimental loop.

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

From a rare IDP state to a practical shortlist, then back to experiment.

## Overview

Design starts with a transient pocket in a rare IDP state.

- Modelling identifies a local pocket that appears only in a rare conformation of the disordered ensemble. Multiple AI approaches can then explore sourceable chemistry against that same structural hypothesis before matched HDX-MS tests the resulting candidates.

## At a glance

- Candidate exploration conditioned on rare, transient IDP conformations
- Commercial building-block constraints for practical synthesis planning
- Multi-model screening across a 12 billion compound search space
- Matched protein-alone and compound-treated HDX-MS measurements
- Protection and exchange kinetics that feed the next design round

## Acts

### Resolve a rare pocket

Ensemble modelling identifies a shallow pocket that exists only when distant segments of the disordered chain make a temporary local contact.

### Explore with multiple models

Different AI models test complementary small-molecule hypotheses against the same ensemble-aware pocket rather than treating one structure as ground truth.

### Constrain to sourceable chemistry

The search is grounded in commercial vendor libraries so selected building blocks can be accessed on a 3 to 4 week timeline.

### Narrow the search

A 12 billion compound search space is filtered to approximately 100 to 300 candidates for experimental prioritisation.

### Return to HDX-MS

Matched protein-alone and compound-treated experiments compare protection and exchange kinetics to evaluate engagement and changes in IDP structural dynamics.

## Experiment decides what advances.

A docking hypothesis earns progression only when matched HDX-MS produces a reproducible protection and kinetics fingerprint.

### Inputs

### Experiment

#### Protection + kinetics

Protein-alone and compound-treated measurements test local protection and exchange behavior under matched conditions.

### Signal

### Model and experiment agree

### Gates

### Engagement

Is local protection reproducible?

A consistent protected region supports direct engagement with the selected IDP state.

### Dynamics

Do exchange kinetics change coherently?

Time-dependent uptake reveals whether the compound changes local ensemble behavior.

### Designability

Can the chemistry be made and improved?

Sourceable building blocks and tractable analogues keep the hypothesis inside a practical design cycle.

### Decisions

### Advance

#### Convergent signal

Reproducible protection, coherent kinetics, and practical chemistry justify the next experimental step.

### Revise

#### Update the hypothesis

Weak, diffuse, or inconsistent evidence redirects the pocket model, chemistry, or experimental design.

### Handoff

### Evidence directs the next design.

Every measured fingerprint becomes a design instruction, either advancing a candidate or refining the hypothesis for the next cycle.