Quantifying the fingerprint descriptor dependence of structure-activity relationship information on a large scale

Area of Science:

• Computational chemistry
• Cheminformatics
• Medicinal chemistry

Background:

• Molecular representations critically influence similarity calculations.
• This representation dependence affects structure-activity relationship (SAR) analysis.
• The extent of SAR information changes due to different molecular representations remains unclear.

Purpose of the Study:

• To quantify the impact of different molecular representations on SAR information.
• To assess how SAR information changes when alternative fingerprints are used as descriptors.
• To determine the proportion of compounds with substantial SAR changes induced by various fingerprints.

Main Methods:

• Quantified SAR information changes across diverse compound classes using six distinct fingerprints.
• Performed SAR profiling on 128 target-based datasets (>60,000 compounds) with high-confidence activity annotations.
• Applied a numerical measure of SAR discontinuity to assess SAR information on a per-compound basis.

Main Results:

• ~70% of compounds exhibited altered SAR characteristics with different molecular representations.
• ~30% of compounds displayed a changed SAR phenotype, indicating distinct fingerprint-dependent local SAR environments.
• Compared fingerprints generated largely incomparable SAR models; atom environment and pharmacophore fingerprints caused the most significant differences.

Conclusions:

• Fingerprint-dependent changes in compound-associated SAR information are substantial and often underestimated.
• The choice of molecular representation significantly impacts SAR analysis outcomes.
• Results highlight the need for careful consideration of molecular representations in drug discovery and SAR studies.