Systematic Data Analysis and Diagnostic Machine Learning Reveal Differences between Compounds with Single- and...

Area of Science:

• Medicinal Chemistry
• Computational Drug Discovery
• Pharmacology

Background:

• Small molecules with multitarget activity, known as promiscuous compounds, are crucial in drug discovery for polypharmacological effects.
• Promiscuity influences drug distribution, pharmacodynamics, and ADMET properties, but features distinguishing single- and multitarget compounds are poorly understood.

Purpose of the Study:

• To identify key features that differentiate single-target from multitarget (promiscuous) compounds.
• To develop accurate predictive models for identifying promiscuous compounds in drug discovery.

Main Methods:

• Systematic data analysis to assemble sets of promiscuous and single-target compounds.
• Machine learning models were employed for prediction of compound promiscuity.
• Molecular similarity analysis and control calculations were used to understand prediction drivers.

Main Results:

• Machine learning accurately predicted promiscuous compounds, primarily driven by structural nearest-neighbor relationships.
• Promiscuous and single-target compounds often formed distinct analog series with unique chemical space coverage.
• Compounds active against functionally distinct targets frequently engaged unique protein targets.

Conclusions:

• Nearest-neighbor effects are critical for predicting compound promiscuity.
• Distinct analog series partitioning underlies structure-promiscuity relationships, providing a rationale for multitarget compound behavior.