Machine learning models for classification tasks related to drug safety

Abstract

In this review, we outline the current trends in the field of machine learning-driven classification studies related to ADME (absorption, distribution, metabolism and excretion) and toxicity endpoints from the past six years (2015–2021). The study focuses only on classification models with large datasets (i.e. more than a thousand compounds). A comprehensive literature search and meta-analysis was carried out for nine different targets: hERG-mediated cardiotoxicity, blood–brain barrier penetration, permeability glycoprotein (P-gp) substrate/inhibitor, cytochrome P450 enzyme family, acute oral toxicity, mutagenicity, carcinogenicity, respiratory toxicity and irritation/corrosion. The comparison of the best classification models was targeted to reveal the differences between machine learning algorithms and modeling types, endpoint-specific performances, dataset sizes and the different validation protocols. Based on the evaluation of the data, we can say that tree-based algorithms are (still) dominating the field, with consensus modeling being an increasing trend in drug safety predictions. Although one can already find classification models with great performances to hERG-mediated cardiotoxicity and the isoenzymes of the cytochrome P450 enzyme family, these targets are still central to ADMET-related research efforts. Graphical abstract

Comparison of Descriptor- and Fingerprint Sets in Machine Learning Models for ADME-Tox Targets

“…Based on the six case studies, we can support our previous analysis of the ADME-Tox-related classification models ( Rácz et al, 2021 ). Here, we have found that the XGBoost algorithm could outperform the neural network-based models, which is consistent with the findings in the mentioned publication: the tree-based algorithms have high priority over the others in the ADME-Tox-related models from the literature of the last 5 years.…”

“…The best benchmark models cannot be determined based on solely one criterion. According to the literature data in our review ( Rácz et al, 2021 ), the present models fit into the average performance in each case study. Moreover, Figure 10C shows that the best of the developed models sometimes can be even better.…”

“…The optimization of the ADME-Tox (adsorption, distribution, metabolism, excretion, and toxicity) properties of potential drug candidates plays a major role in any drug design project. Drug candidates often fail due to their suboptimal drug safety properties ( Ferreira and Andricopulo, 2019 ; Rácz et al, 2021 ; Danishuddin et al, 2022 ). As the in vivo ADME-Tox experiments are expensive and time-consuming, in silico QSPR models are frequently used for the prediction of ADME-Tox properties with the great advantage as they can be used before the synthesis of the compounds ( Yang et al, 2018 ).…”

“…In our recent review, a detailed comparison of machine learning–driven ADME-Tox classification models was presented concentrated on the last 5 years. Our analysis showed that tree-based methods are the most popular choices amongst the machine learning algorithms for ADME-Tox model development ( Rácz et al, 2021 ). The superiority of the machine learning algorithms over the traditional classification methods was also mentioned in the work of Tsou et al (2020) .…”

Quantitative Retention (Structure)–Activity Relationships in Predicting the Pharmaceutical and Toxic Properties of Potential Pesticides

“…The physicochemical, pharmacokinetic, and toxicity parameters of the compounds ( Table 2 ) are as follows: the logarithm of the partition coefficient (log P ) in the n -octanol/water system, the number of hydrogen bond donors ( HBD ), acceptors ( HBA ), and rotatable bonds ( NRB ), molar weight ( MW ), topological polar surface area ( TPSA ) [ 24 ], the volume of distribution in the body ( V d ) [ 25 ], the fraction unbonded in a brain ( f u, brain ), in plasma ( f u, plasma ) [ 26 ], and pharmacokinetic parameters describing blood–brain distribution (log BB ) [ 26 , 27 , 28 , 29 ], the rate of permeation from aqueous solutions through skin (log K p ) [ 30 , 31 ], skin–water partition coefficient (log K sc ) describing dermal absorption from aqueous solutions [ 32 , 33 ], the rate of permeation through cell (log K w/cell ) [ 34 ], partitioning between water and serum albumin (log P w/HSA ), and binding to human serum albumin (log K HSA ) [ 10 , 35 , 36 , 37 ], partitioning between water and plant’ cuticles (log P w/pc ) [ 38 ], and the dose causing the death of 50% of the group of mice tested after oral administration ( LD 50 ) [ 39 , 40 ]. These parameters describe important properties of the test substances and provide information about their potential applications as pesticides as well as potential threats to humans [ 41 , 42 ].…”