Vivofold Research

Structural Bioinformatics & Computational Drug Discovery

Computational drug discovery produces many confident numbers and fewer reliable conclusions. Our work emphasises validation: retrospective benchmarks for docking protocols, leakage-free QSAR evaluation and applicability domains, and clear reporting of what a prediction can support.

2D protein–ligand interaction diagram of a docked pose: hydrogen bonds (dashed, with donor–acceptor distances) and hydrophobic contacts (dotted).

Questions

What we want to know

  1. How often do docking protocols pass retrospective validation for the targets they are applied to?
  2. How much do random splits overstate QSAR performance compared with cluster or scaffold splits?
  3. Which structural features of predicted models matter most for docking reliability?

Approach

How we work on it

  • Redocking and decoy-enrichment benchmarks
  • Cluster-aware QSAR validation with Y-randomisation
  • Comparison of experimental and predicted structures

Methods already implemented

Random forest, kNN and ridge QSAR with applicability domainsButina clustering of chemical spaceInteraction profiling and pocket detectionKabsch superposition and TM-score

Open projects

Projects you can join

01

Docking validation audit

Systematically test docking protocols on public targets with known actives and decoys.

02

QSAR split study

Quantify performance inflation from random splits across ChEMBL targets.

03

Predicted structures for docking

Compare docking outcomes on experimental and predicted structures of the same targets.

Collaborate

Propose a project in structural bioinformatics & computational drug discovery

Sent securely to [email protected]. Your details are used only to process this request.