Self-organizing molecular fingerprints: a ligand-based view on drug-like chemical space and off-target prediction

Background: Reliable prediction of multiple ligand receptor interactions for a given bioactive compound helps recognize and understand off-target effects, and enables drug re-purposing and scaffold-hopping in lead discovery. We developed a ligand-based computational method for drug-target prediction that is independent from protein structural analysis. Method: The idea is to infer drug targets from the pharmacophoric feature similarity of known ligands, and define functional target similarity from a ligand perspective, which also provides access to targets with unknown structures. First, known ligands were represented by topological pharmacophoric features. Then, the self-organizing map technique was used to generate fingerprint patterns for similarity analysis, where each resulting fingerprint represents a drug target. Target fingerprints were clustered and analyzed for correlations. Well-structured dendrograms were obtained presenting interpretable and meaningful relationships between drug targets. Conclusion: Self-organization of fingerprints reduces noise from molecular pharmacophore descriptors, captures their essential features, and reveals potential cross-activities of ligand classes and off-target effects of bioactive compounds.