Do Molecular Fingerprints Identify Diverse Active Drugs in Large-Scale Virtual Screening? (No).

Area of Science:

• Computational chemistry and cheminformatics
• Small-molecule drug discovery

Background:

• Large-scale virtual screening requires efficient molecular representations.
• Molecular fingerprints are commonly used for representing molecules and predicting properties.
• The effectiveness of fingerprints in identifying biologically active molecules needs further investigation.

Purpose of the Study:

• To evaluate the utility of standard molecular fingerprints for predicting similar molecular activity.
• To assess the discriminative power of fingerprint similarity in distinguishing active from inactive compounds.
• To explore the relationship between fingerprint similarity and compound potency.

Main Methods:

• Analysis of commonly used molecular fingerprints.
• Assessment of fingerprint similarity in predicting activity against a target protein.
• Evaluation of enrichment and potency correlation in screened datasets.

Main Results:

• Fingerprint similarity shows low discriminative power between active and inactive molecules.
• While some enrichment of active molecules is observed, datasets remain dominated by inactive compounds.
• High-similarity active molecules often share structural scaffolds, suggesting enumeration as an alternative.
• Fingerprint similarity does not correlate with compound potency, even among active molecules.

Conclusions:

• Standard molecular fingerprints are insufficient for reliably predicting similar molecular activity and potency.
• The findings underscore the limitations of current fingerprint-based approaches in large-scale drug discovery.
• Development of novel molecular representations is crucial for enhancing the identification of biologically active molecules.