Concept of Maximum Drug Likeness
Traditional virtual screening often optimizes for activity or binding alone, which contributes to late‑stage attrition. Maximum Drug Likeness (MDL) reframes screening around the holistic similarity of a candidate to real, approved drugs. Fivefold MDL (5F‑MDL) quantifies that similarity across five critical dimensions:
(1) Physicochemical: properties that define the developability window (e.g., size, polarity, solubility, permeability).
(2) Pharmacokinetics (ADME): absorption, distribution, metabolism, clearance, half‑life.
(3) Efficacy: potency and mechanism‑linked readouts (e.g., IC50/EC50/MIC, target engagement, exposure–response).
(4) Safety: cytotoxicity and system liabilities (e.g., genotoxic, cardiac, hepatic/renal, DDI propensity).
(5) Stability: chemical/physical stability from bench to shelf and in biological matrices.
Operationally, 5F‑MDL uses 33 deep‑learning QSAR sub‑models to generate a 33‑dimensional “property spectrum” for any molecule. After scale harmonization and probability mapping, we compute an S5F score that measures how closely a candidate’s spectrum aligns with those of reference drugs; higher S5F indicates stronger multi‑dimensional drug likeness.
