Science

The models, databases and methods behind ProteinInsight — and their limits. Every result is research-use only and labelled with its provenance.

Purpose-built for the entire drug-development journey

Target discovery to post-market — twelve stages in one coherent workspace, not generic project tooling.

01Disease & Target02Candidate Discovery03Optimization04Preclinical05IND06Phase 107Phase 208Phase 309NDA / BLA10CMC / GMP11Launch12Post-market

Methods & models

What each analysis computes, and the confidence you should place in it.

Structure prediction

ESMFold for single sequences (public API, up to ~400 residues) and AlphaFold DB for any length by UniProt accession. Each residue carries a pLDDT confidence (0–100); low-pLDDT regions are unreliable.

Molecular docking

Poses a ligand in a target's binding pocket and scores the interaction. Scores rank hypotheses for triage — they are not binding-affinity measurements.

ADMET prediction

Rule- and descriptor-based estimates of absorption, distribution, metabolism, excretion and toxicity risk to flag liabilities early.

PK/PD simulation

Compartmental pharmacokinetic exposure and pharmacodynamic response curves under configurable dosing, for what-if exploration.

Safety signals

Adverse-event mining from openFDA FAERS and bioactivity from ChEMBL. Post-market reports are signals, not proof of causation.

Data sources

External data is fetched live from pinned public databases and cached for interpretation.

UniProt

Protein sequences, function, and cross-references, resolved by accession.

RCSB PDB

Experimentally determined 3D structures.

AlphaFold DB

Precomputed AlphaFold predicted structures, any length.

ESM Atlas

ESMFold single-sequence structure prediction.

ChEMBL

Curated bioactivity and medicinal-chemistry data.

openFDA

FAERS adverse-event reports for post-market safety.

ProteinInsight is a research tool. Nothing here is validated for clinical, diagnostic, or regulatory use, and predictions must be confirmed experimentally.