Comparative Study of Multitask Toxicity Modeling on a Broad Chemical Space

Abstract

Acute toxicity is one of the most challenging properties to predict purely with computational methods due to its direct relationship to biological interactions. Moreover, toxicity can be represented by different endpoints: it can be measured for different species using different types of administration, etc., and it is questionable if the knowledge transfer between endpoints is possible. We performed a comparative study of prediction multi-task toxicity for a broad chemical space using different descriptors and modeling algorithms and applied multi-task learning for a large toxicity dataset extracted from the Registry of Toxic Effects of Chemical Substances (RTECS). We demonstrated that multi-task modeling provides significant improvement over singleoutput models and other machine learning methods. Our research reveals that multi-task learning can be very useful to improve the quality of acute toxicity modeling and raises a discussion about the usage of multi-task approaches for regulation purposes.

Could Graph Neural Networks Learn Better Molecular Representation for Drug Discovery? A Comparison Study of Descriptor-based and Graph-based Models

“…Moreover, it can be observed that the top three models of all the datasets were mainly occupied by the descriptorbased models (the ratio is 24/33=73%), substantiating the more powerful predictive abilities of the descriptor-based models compared to the graph-based models. Here what we found is that the graph-based models can outperform the descriptor-based models on some lager or multi-task datasets such as the HIV, Tox21 and ToxCast datasets, which is well accord with the previous conclusions where DNN excel at larger amounts of data and multi-task learning [65,66]. However, to build such generalizable and robust deep models requires large-scale high-quality datasets and the datasets in the practical drug discovery campaigns routinely suffer from narrow chemical diversity and insignificant sample sizes [67].…”

“…Numerous studies demonstrated that multi-task models have advantages over single-task models due to their ability to excavate the inconspicuous hidden relations between different subtasks and transparently share the learned features among all the tasks. [58,65,66] Nevertheless, the performance of multi-task models is highly related to the favorable correlations of individual tasks but such ready-to-use tasks are not so commonly seen in practical drug discovery campaigns.…”

Transformer-CNN: Swiss knife for QSAR modeling and interpretation

“…Sml2canSml was added as CDDD descriptors to OCHEM. These descriptors were analysed by the same methods as used in the previous work, i.e., LibSVM [57], Random Forest [58], XGBoost [59] as well as by Associative Neural Networks (ASNN) [60] and Deep Neural Networks [61]. Exactly the same protocol, fivefold cross-validation, was used for all calculations.…”

Analysis and Modelling of False Positives in GPCR Assays

“…Their detailed description can be found elsewhere [4]. Associative Neural Networks (ASNN) [11], Deep Neural Network (DNN) [12], Extreme Gradient Boost (XGBOOST) [13], and Least Squares Support Vector Machine (LSSVM) [14] algorithms were analyzed for training the models. The methods were used with default parameters as specified on the OCHEM web site.…”