MAP4

MAP4 encodes molecules as MinHashed atom-pair fingerprints up to a diameter of four bonds to enable similarity searches and chemical-space mapping across both small molecules and large biomolecules.

Key Features:

• Hybrid Fingerprint Design: Combines substructure and atom-pair concepts by capturing circular substructures with radii of 1 and 2 bonds around each atom in an atom-pair, represented as pairs of SMILES strings combined with topological distances.
• MinHashing Technique: Atom-pair molecular shingles are hashed and MinHashed to form the MAP4 fingerprint, reducing dimensionality and improving computational efficiency while preserving similarity information.
• Performance Superiority: Outperforms existing fingerprints on an extended benchmark including Riniker and Landrum small-molecule benchmarks and peptide benchmarks, and effectively recovers BLAST analogs from scrambled or point-mutation analogs.
• Chemical Space Mapping: Generates well-organized tree-maps (TMAPs) for databases such as DrugBank, ChEMBL, SwissProt, and the Human Metabolome Database (HMBD), and distinguishes metabolites in HMBD where over 70% are indistinguishable using substructure fingerprints alone.

Scientific Applications:

• Virtual Screening: Facilitates similarity-based virtual screening across diverse chemical spaces by representing both small molecules and large biomolecules with a single fingerprint.
• Chemical Space Exploration: Enables chemical-space visualization and clustering of entries from databases like DrugBank and ChEMBL using TMAPs derived from MAP4 fingerprints.
• Metabolomics Research: Supports differentiation and characterization of closely related metabolites in the Human Metabolome Database (HMBD), aiding metabolite identification and pathway analysis.

Methodology:

MAP4 creates atom-pair molecular shingles by considering circular substructures of radii 1 and 2 bonds around each atom pair (represented as SMILES pairs with topological distances), and these shingles are hashed and MinHashed to produce the compact MAP4 fingerprint.