Combinatorial Perturbation Prediction Using Causally-Inspired Neural Networks

Explore a groundbreaking approach to phenotype-driven drug discovery in this hour-long conference talk. Delve into PDGRAPHER, a causally-inspired graph neural network model designed to predict perturbagens capable of reversing disease effects. Learn how this innovative method expands the search space for new therapeutic agents by directly predicting perturbagens, offering a faster and more comprehensive alternative to traditional approaches. Discover the model's performance across eight datasets of genetic and chemical perturbations, its ability to rank therapeutic targets, and its significant improvements in training efficiency. Gain insights into structural causal models, dataset preparation, training processes, and key results. Conclude with valuable takeaways and participate in a Q&A session to deepen your understanding of this cutting-edge research in AI-driven drug discovery.

- Intro + Background
- Structural Causal Models
- PDGrapher
- Datasets + Training
- Results
- Takeaways
- Q&A