Chemical rules for optimization of chemical mutagenicity via matched molecular pairs analysis and machine learning methods

Abstract

Chemical mutagenicity is a serious issue that needs to be addressed in early drug discovery. Over a long period of time, medicinal chemists have manually summarized a series of empirical rules for the optimization of chemical mutagenicity. However, given the rising amount of data, it is getting more difficult for medicinal chemists to identify more comprehensive chemical rules behind the biochemical data. Herein, we integrated a large Ames mutagenicity data set with 8576 compounds to derive mutagenicity transformation rules for reversing Ames mutagenicity via matched molecular pairs analysis. A well-trained consensus model with a reasonable applicability domain was constructed, which showed favorable performance in the external validation set with an accuracy of 0.815. The model was used to assess the generalizability and validity of these mutagenicity transformation rules. The results demonstrated that these rules were of great value and could provide inspiration for the structural modifications of compounds with potential mutagenic effects. We also found that the local chemical environment of the attachment points of rules was critical for successful transformation. To facilitate the use of these mutagenicity transformation rules, we integrated them into ADMETopt2 ( http://lmmd.ecust.edu.cn/admetsar2/admetopt2/ ), a free web server for optimization of chemical ADMET properties. The above-mentioned approach would be extended to the optimization of other toxicity endpoints.

Keywords: Consensus model; Lead optimization; Machine learning; Matched molecular pairs analysis; Mutagenicity optimization.

Fig. 1

The workflow of this study includes four steps: a data collection and preparation, b matched molecular pairs analysis to derive mutagenicity transformation rules, c the construction of machine learning models for mutagenicity prediction, and d evaluation of mutagenicity transformation rules via machine learning models

Fig. 2

The heat map a and the molecular cloud b of the Ames data set

Fig. 3

The model performance of six base classifiers (GNN model, RF_RDK model, SVM_ECFP model, LGB_RDK model, XGB_MACCS model and GBT_MACCS model) and consensus model in the test set a and external validation set b. The ‘Applicability Domain’ referred to the performance of consensus model considering only the compounds within the applicability domain

Fig. 4

The distribution of the prediction results of newly generated compounds from the DGM/NIHS data set a and Hansen/ISSSTY data set b

Fig. 5

The influence of rule frequency on transformation validity. The S_t.v. were calculated through DGM/NIHS data set a and Hansen/ISSSTY data set b, respectively

Fig. 6

The structures of nifurtimox (compound A) and metronidazole (compound B), and the newly generated compounds (compounds A1, A2, and B1, B2 from the optimization of nifurtimox and metronidazole, respectively). The compounds in red boxes were predicted as Ames positive, and the ones in green boxes were predicted as Ames negative