From Static Snapshots to Dynamic Ensembles
Transforming protein structure prediction from static snapshots to dynamic ensembles for de novo design and drug discovery.
The Breathing Protein
Proteins are not static statues - they vibrate, shift, and breathe. Hover the visualization to trigger a conformational shift from a closed (inactive) to an open (active) state.
Limitations of rigid-body assumptions
Conventional algorithms rely on static crystal structures, ignoring thermodynamic fluctuations.
This leads to false negatives in mutation-impact analysis, as steric clashes and entropy changes are overlooked.
Enhanced sampling & ensemble generation
We integrate deep-learning potentials with molecular dynamics to sample the conformational landscape.
Boltzmann-weighted ensembles capture transient states critical for function - not just the lowest-energy snapshot.
Cryptic pocket identification & ligand docking
We identify cryptic allosteric sites that only open in specific conformational states.
This enables rational small-molecule design for 'undruggable' targets and de novo enzyme engineering.
CASP blind prediction results
Consistent top performance at the most rigorous international benchmarks in structural bioinformatics.
Top-5 refinement method
CASP13 refinement category
Best Server award
Lifts models beyond AlphaFold2 accuracy
2nd place · ligand-pose
Surpasses AlphaFold3 server
Outperforms AlphaFold3
Quaternary structure prediction
30,000+ users
400,000+ jobs processed per year
16,000+ engineered features
Protein binder quality assessment
From research to cancer vaccines
Currently building a TCR-pMHC structural prediction pipeline for the personalised cancer vaccine programme at InstaDeep (BioNTech Group), integrating AlphaFold2, Boltz-2, and Chai-1 to inform neoantigen targeting.