Poster

Using Agentic AI systems to harness multi-omics foundation models for accelerating biomedical discovery

Abstract

Emerging AI-based technologies such as Foundation Models (FMs) and Agentic AI, have revolutionised AI and provide opportunities to accelerate biomedical discovery. We harness such technology to identify genetic and molecular targets for disease and treatment development. Our family of Biomedical FMs span molecular domains—including small molecule and biologics discovery—and omics research. These models are pre-trained on diverse patient data, including genomics, transcriptomics and molecular-level data such as small molecules and proteomics. Using multi-omics allows us to offer a more comprehensive understanding of the complex interactions between disease mechanisms and individual variability. Our models are validated on markers and models of disease and allow simulation of interventions such as drug treatment, to prioritise experimentation candidates. Some of our models are open-sourced such as mammal, and mmelon.

Our FMs can be used for inference directly for a range of tasks, or they can be fine-tuned with additional datasets to broaden their applicability to new downstream tasks. In either case, construction and execution of a FM workflow can be a time consuming and interdisciplinary task, confining usage to specialised computational experts. For example, workflows move from processing raw omics or molecular data, through FM inference, to interpretation of the biological validity of the results. Agentic AI are LLM-driven systems which autonomously plan, reason and dynamically call tools/functions. They are powerful for planning and executing complex workflows, removing coding and computational pre-requisites. As such, we bring our FMs into Agentic AI systems, capable of complex omics workflow construction and execution and equipped with domain specific tools and a ‘domain-expert’, a concept which incorporates domain specific knowledge, guiding the agent to more accurately plan omics’ workflows and prevent LLM inaccuracies. We hope to show that Agentic AI systems can enhance omics analysis in terms of scalability, reproducibility and scientific accuracy.

What sets us apart is our combination of the following: (1) cutting-edge AI-based technologies (2) multi-modal pre-training datasets (3) an interdisciplinary research team including multi-omics specialists, biologists, computational biologists, bioinformaticians, mathematicians, computational scientists and AI specialists and (4) our end-user translational focus that provides a testbed for AI application. Globally across IBM Research our partner is the Cleveland Clinic where we collaboratively drive advances in healthcare and treatment target discovery. In the UK we work with industrial partners focused on healthcare and life sciences within a unique collaborative centre led by IBM Research and STFC, called the Hartree National Centre for Digital Innovation.

Related