Structural Analysis and Prediction of Hematotoxicity Using Deep Learning Approaches

Abstract

Hematotoxicity has been becoming a serious but overlooked toxicity in drug discovery. However, only a few in silico models have been reported for the prediction of hematotoxicity. In this study, we constructed a high-quality dataset comprising 759 hematotoxic compounds and 1623 nonhematotoxic compounds and then established a series of classification models based on a combination of seven machine learning (ML) algorithms and nine molecular representations. The results based on two data partitioning strategies and applicability domain (AD) analysis illustrate that the best prediction model based on Attentive FP yielded a balanced accuracy (BA) of 72.6%, an area under the receiver operating characteristic curve (AUC) value of 76.8% for the validation set, and a BA of 69.2%, an AUC of 75.9% for the test set. In addition, compared with existing filtering rules and models, our model achieved the highest BA value of 67.5% for the external validation set. Additionally, the shapley additive explanation (SHAP) and atom heatmap approaches were utilized to discover the important features and structural fragments related to hematotoxicity, which could offer helpful tips to detect undesired positive substances. Furthermore, matched molecular pair analysis (MMPA) and representative substructure derivation technique were employed to further characterize and investigate the transformation principles and distinctive structural features of hematotoxic chemicals. We believe that the novel graph-based deep learning algorithms and insightful interpretation presented in this study can be used as a trustworthy and effective tool to assess hematotoxicity in the development of new drugs.

In Silico Prediction of Chemical Acute Dermal Toxicity Using Explainable Machine Learning Methods

“…In this context, SARpy software was employed to extract SAs, and the accuracy was calculated for SA analysis. We established specific thresholds for SA extraction in both data sets, wherein the occurrence frequency ( N P ) of a SA in toxic compounds had to be greater than or equal to 8, and the accuracy, defined as the ratio of toxic compounds containing the SA to all compounds containing the SA, had to be higher than 0.70 . In the end, we selected 10 SAs for each data set, which are presented in Tables and .…”

“…DL has been increasingly applied in various research domains. While in many instances, DL models outperform traditional ML models in compound property prediction, , it is important to note that when the data set is limited in size, ML often yields better results than DL. ,, Furthermore, DL algorithms are considered to perform well on large data sets, as they can learn more complex features from the training set . In the models we constructed, due to the use of undersampling methods, the modeling data rely on a limited number of samples.…”

Rapid Screening of Chemicals with Placental Transfer Risk Using Interpretable Machine Learning

“…We further used the SHAP method to interpret the result analyzed by the retrained LSTM model . Among all fingerprints, MACCS_keys_86 had the strongest predictive value for our model, followed quite closely by MACCS_keys_53.…”

In silico prediction of ocular toxicity of compounds using explainable machine learning and deep learning approaches

“…Furthermore, the chemical diversity of a group of molecules can be confirmed by comparing their molecular scaffolds (Bajorath, 2018). In this study, the chemical scaffold analysis was conducted using the Murcko scaffold (Long et al, 2023), which represents the only ligand and ring system remaining after removing all substituents. Derived from the Murcko scaffold, the carbon backbone is generated by converting all heteroatoms into carbon atoms and all bonds into single bonds (Bemis & Mark, 1996).…”