A bioinformatics platform for prioritizing protein targets in pathogens.

Proteins are assessed for selectivity against the human proteome and gut microbiome, and for essentiality using the Database of Essential Genes. Subcellular localization is predicted by PSORTb to evaluate target accessibility.

Structures are assigned by priority: experimental PDB entries first, then AlphaFold Database models, then ColabFold predictions for proteins without other coverage. Per-residue pLDDT scores indicate model confidence.

FPocket detects surface cavities using Voronoi spheres and reports a druggability score from 0 to 1. P2Rank applies a random forest trained on ~15,000 PDB structures to rank predicted binding site probability.

InterProScan scans each protein against 20 databases including Pfam, HAMAP, and PANTHER to assign domains, GO terms, and EC numbers. UniProt provides additional curated GO and enzyme data for mapped accessions.

LigQ_2 is the internal ligand-evidence step. It retrieves co-crystallized ligands from PDB, measured bioactive compounds from ChEMBL, and proposed ZINC compounds by chemical similarity. Direct and homology-transferred evidence are distinguished.

Physicochemical properties including MW, LogP, and TPSA are computed from SMILES using RDKit. Lipinski Rule of Five compliance and PAINS substructure filters are applied to flag compounds prone to assay interference.