Background
The ability to predict how human cells will respond to a genetic change or a new drug is a grand challenge in medicine. While we can generate vast amounts of biological data, interpreting its complex, networked structure to find the true causal drivers of disease remains a major hurdle. This project will develop a novel Artificial Intelligence system to build an interpretable, predictive model of cellular biology.
Novelty & Importance
We will use Geometric Deep Learning, a cutting-edge AI technique, to learn from a massive biological Knowledge Graph that integrates data from genomics, proteomics, and cell biology. The key innovation is to engineer a Graph Neural Network (GNN) that functions as an in silico perturbation engine. This moves beyond simple statistical prediction to create a system that can simulate the causal consequences of modifying a gene or introducing a drug, allowing us to systematically test molecular hypotheses at an unprecedented scale.
Aims & Objectives
The primary aim is to engineer a robust and interpretable AI engine for causal inference in biology. The student will first develop the data pipelines to build the foundational knowledge graph. They will then implement and train the GNN, using large-scale human genetic data as a supervisory signal to learn the fundamental rules of the biological network. Finally, they will build and validate the perturbation engine, using it to identify optimal intervention points in disease networks and generate human-interpretable mechanistic hypotheses, paving the way for a new generation of AI-driven therapeutic discovery.