QAFFP

Quantitative Affinity Fingerprints (QAFFP): Bioactivity-Based Molecular Encoding Framework

Quantitative Affinity Fingerprints (QAFFP) encode small molecules in bioactivity space by aggregating predicted activity profiles from QSAR models rather than relying on chemical structure similarity.

Key Features:

• Bioactivity Encoding: Encodes compounds using similarity in predicted bioactivity profiles, integrating activity data from structurally dissimilar molecules to characterize compound behavior.
• QSAR Model Integration: Applies 1360 QSAR models derived from ChEMBL binding affinity (Ki, Kd) and inhibitory concentration (IC50, EC50) data to generate affinity fingerprints across diverse assays.
• Benchmarking and Validation: Evaluates predictive performance using IC50 data from 18 cancer cell lines and 25 protein target datasets from ChEMBL.
• Predictive Performance: Achieves prediction errors of approximately 0.65–0.95 pIC50 units on test sets; slightly lower accuracy than Morgan2 fingerprints and physicochemical descriptors (effect size 0.02–0.08 pIC50 units).

Scientific Applications:

• Drug Discovery: Supports bioactivity classification, similarity searching, scaffold hopping, therapeutic candidate identification, and prediction of off-target effects or toxicity based on bioactivity profiles.

Methodology:

Generates quantitative affinity fingerprints by applying QSAR models trained on Ki, Kd, IC50, and EC50 data from ChEMBL to predict compound bioactivity profiles. These predicted profiles function as descriptors for modeling activities across assays and enable comparison of compounds with limited structural similarity.