Assessing Neurotoxicity Potential of Chemicals Through Adverse Outcome Pathways: A Multitask Learning Approach

Vol 2, 2022 - 153233
Poster
Favorite this paper
How to cite this paper?
Abstract

Developmental neurotoxicity (DNT) is an important endpoint for the safety assessment of chemicals. However, the current in vivo and in vitro models for DNT assessment are expensive and may not fully capture relevant pathway interference mechanisms to neurotoxicity in humans.1 In this sense, we developed a reliable, time- and cost-efficient multitask learning (MTL) approach to predict DNT-molecular initiating events (MIEs) from adverse outcome pathways. Initially, a sparse matrix composed of 24,900 compounds vs. 19 MIEs (AChE, GABA-A, D2, NMDAR, etc.) was collected e curated from the ChEMBL30. In parallel, a task relevance analysis indicated that compounds from different endpoints share the same chemical space. Then, MTL models were developed using Feedforward Neural Networks (FNNs) and graph-based Message Passing Neural Networks (MPNNs) architectures. Compared with single-task learning (STL) approaches focused only on a single endpoint, our best MTL model (FNN+ECFP4 fingerprints) significantly outperforms past methods (average RMSE ~0.6, average r ~0.8).

Share your ideas or questions with the authors!

Did you know that the greatest stimulus in scientific and cultural development is curiosity? Leave your questions or suggestions to the author!

Sign in to interact

Have a question or suggestion? Share your feedback with the authors!

Institutions
  • 1 Universidade Federal de Goiás
Track
  • 2. Medicinal Chemistry of Synthetic Compounds
Keywords
computational toxicology
Transfer Learning
Molecular Initiating Events
Risk Assessment