Megha Hegde

Research project: Deciphering Multi-omics Data: Large Language Models in Disease Research

Abstract

This project uses artificial intelligence (AI) to support disease-related research by designing novel large language models (LLMs) that can decipher multi-omics data.The project aims to go beyond current Transformer models by developing a novel unsupervised pre-training paradigm, which can better model the underlying "language" of multi-omics data. Additionally, as the attention mechanism has proven to be a bottleneck for genomic modelling due to its quadratic complexity, we explore alternatives that can provide excellent performance with a lower computational cost. A key area of focus is modelling long-range dependencies within the data, which is crucial in identifying patterns (for instance, drug-resistant mutation profiles) that are currently out of reach for state-of-the-art models.This research has potential applications to a wide range of communicable diseases caused by pathogens, including HIV, tuberculosis bacteria, MRSA, and hepatitis B virus, which have a significant burden on human health worldwide and have become resistant to many drugs used to treat them. The developed methods can also support cancer research by integrating various data types and sources, particularly in developing patient-specific treatment regimens.