Poster: Gaining Insight into Toxicity Predicting Machine Learning Algorithms

Poster presentation from our collaboration with the University of Cambridge. Neural networks have been constructed for the prediction of important human Molecular Initiating Events (MIEs) for use in safety assessment. Open source data from ChEMBL and ToxCast was used, providing a balance of positive and negative data points for several human MIEs, including G-protein coupled receptors, nuclear receptors, enzymes, ion channels and transporters. These networks show extremely high performance (accuracy >90% in most cases), as expected, and a similarity algorithm has been developed to assess how the signal in the network propagates through it when a chemical is introduced. This allows the model to provide activity predictions for new chemicals and training set molecules with high network similarity, meaning the prediction can be treated in a read-across style manner by the user, increasing their confidence in the computer’s prediction.

This poster was presented by Tim Allen (University of Cambridge).

View the poster here.


Comments are closed.