Computing similarity between structural environments of mutagenicity alerts.

Area of Science:

• Computational chemistry
• Toxicology
• cheminformatics

Background:

• Accurate prediction of chemical mutagenicity is crucial for drug discovery and safety assessment.
• Existing methods for identifying mutagenic alerts and their analogues can be limited in accuracy and scope.

Purpose of the Study:

• To develop and validate a novel method for generating molecular fingerprints from structural environments of mutagenicity alerts.
• To enhance the classification accuracy of mutagenicity alerts and facilitate the search for structurally similar chemical analogues.
• To explore the utility of transfer learning for improving fingerprint generation and similarity searches on smaller, labelled datasets.

Main Methods:

• Generating molecular fingerprints based on the structural environments surrounding known mutagenicity alerts.
• Calculating chemical similarity between these generated fingerprints.
• Applying transfer learning by pre-training chemical fragment representations on large unlabeled datasets before applying them to Ames test outcome data.
• Utilizing a k-nearest neighbors prediction method and analyzing the impact of hyperparameters like similarity thresholds and alert environment size.

Main Results:

• The developed approach improved classification accuracy for mutagenicity alerts.
• Distributed fingerprints generated using transfer learning demonstrated superior prediction performance and increased coverage compared to traditional binary fingerprints.
• The methodology was successfully applied to four common mutagenic functionalities: primary aromatic amine, aromatic nitro, epoxide, and alkyl chloride.
• The study detailed the effects of various hyperparameters on prediction accuracy and test coverage.

Conclusions:

• The novel method for generating molecular fingerprints from alert environments effectively improves mutagenicity prediction and analogue searching.
• Transfer learning significantly enhances the performance of chemical fragment representations for mutagenicity assessment, especially with limited labelled data.
• The findings provide a valuable tool for computational toxicology and cheminformatics, aiding in the identification and assessment of potentially mutagenic compounds.