BenevolentAl Therapeutics Pipeline and Triage

Target Prediction Many strands of Al are used to generate target predictions Knowledge Graph 586 Path-based Interpretability Tensor Factorization Human Queries Accuracy Hypotheses Augmentation Learned Weighted Rank Aggregation Multi-prediction Aggregation Specialized models are designed to best produce complementary predictions from the Benevolent Knowledge Graph. Smart aggregation combines predictions by examining which models have been successful in the past, learning what combination of features - and which kinds of evidence - identify a successful target. Al-derived assistance provides support for predictions, for example, from Graph Convolutional Neural Networks. 'Neil et al, 2018 - https://arxiv.org/abs/1812.00279 Data-modality specific Genetic Basis Results in a list of potential therapeutic targets for expert triage Benevolent 19

View entire presentation