Machine Learning-Based Models with High Accuracy and Broad Applicability Domains for Screening PMT/vPvM Substances

Abstract

Persistent, mobile, and toxic (PMT) substances and very persistent and very mobile (vPvM) substances can transport over long distances from various sources, increasing the public health risk. A rapid and high-throughput screening of PMT/vPvM substances is thus warranted to the risk prevention and mitigation measures. Herein, we construct a machine learning-based screening system integrated with five models for high-throughput classification of PMT/vPvM substances. The models are constructed with 44 971 substances by conventional learning, deep learning, and ensemble learning algorithms, among which, LightGBM and XGBoost outperform other algorithms with metrics exceeding 0.900. Good model interpretability is achieved through the number of free halogen atoms (fr_halogen) and the logarithm of partition coefficient (MolLogP) as the two most critical molecular descriptors representing the persistence and mobility of substances, respectively. Our screening system exhibits a great generalization capability with area under the receiver operating characteristic curve (AUROC) above 0.951 and is successfully applied to the persistent organic pollutants (POPs), prioritized PMT/vPvM substances, and pesticides. The screening system constructed in this study can serve as an efficient and reliable tool for high-throughput risk assessment and the prioritization of managing emerging contaminants.

Machine Learning Model for Screening Thyroid Stimulating Hormone Receptor Agonists Based on Updated Datasets and Improved Applicability Domain Metrics

“…The broad chemical space of the dataset is critical for the ML modeling and can widen AD of the classifiers, as can also be concluded from a previous study. 18 The number of compounds in most of the communities also increased after the data incorporation. It can be perceived that the more the communities in a dataset, the more the compounds in communities, the more the ML models can learn, and the higher performance the models have.…”

“…For the 22 classifiers, the differences (δ) between A ROC on the training set and those on the test set range from 0.027 to 0.076 (Table S3). It can be concluded from previous studies that if δ/ A ROC ≤ 10% ( A ROC on the training set), the corresponding classifiers are free of overfitting. ,, For the 22 classifiers, the δ/ A ROC values are all ≤8%. Therefore, there is scarcely any overfitting in the constructed ML classifiers.…”

“…set), the corresponding classifiers are free of overfitting. 18,57,58 For the 22 classifiers, the δ/A ROC values are all ≤8%. Therefore, there is scarcely any overfitting in the constructed ML classifiers.…”

“…Quantitative structure−activity relationship (QSAR) models, as well-established representatives in computational toxicology, 17,18 have been proven to be successful in screening hazardous chemicals. 19−21 There have been previous efforts to develop machine learning (ML)-based QSAR models for screening TSHR agonists.…”

“…Quantitative structure–activity relationship (QSAR) models, as well-established representatives in computational toxicology, , have been proven to be successful in screening hazardous chemicals. − There have been previous efforts to develop machine learning (ML)-based QSAR models for screening TSHR agonists. − However, the previous models were trained with either small datasets or severely imbalanced datasets of classifications. It is known that ML-based QSAR models built on small datasets can be unreliable, and the learned knowledge is fragile .…”