Explore the integration of diverse data modalities, including genomics, clinical records, and imaging, using machine learning for a comprehensive understanding of complex diseases in preclinical and clinical contexts. While traditional multiomics analysis combines RNA, proteomics, and clinical data, this presentation highlights the innovative inclusion of imaging data, and enriching the multi-modal framework. Learn about practical applications of multiomics analysis, combining imaging and genomics data, and delve into exciting developments in radiogenomics and pathogenomics for a holistic understanding of disease mechanisms.

Genomics England and its partners are currently building the world's largest multimodal research platform for cancer. The platform uniquely brings together clinical WGS and healthcare data—from Genomics England's work enabling the NHS to deliver precision medicine through the Genomic Medicine Service—with hundreds of thousands of matched pathology and radiology images being digitized by the National Pathology Imaging Consortium and made accessible through AWS. Join us to hear about the progress and challenges of making 100PB of data queriable in the cloud; a first look at the sort of insights this platform can provide; and how industry can use it and partner with Genomics England and its participants to advance new therapies.

Large Language Models (LLMs) represent a transformational technology. Despite their prowess, they face limitations in handling complex data types, like genomics, and in citing references or incorporating the latest findings effectively. The integration of LLMs with continually updated knowledge bases can create conversational AI tools that not only augment and amplify the reasoning and judgment of experts but also ensure accurate and relevant insights. This presentation will cover core principles, proven patterns, and common challenges with deploying practical and trustworthy AI Assistants in regulated and rapidly evolving industries.

Lung cancer stands as a significant global health challenge, with early diagnosis playing a pivotal role in improving patient outcomes. Current diagnostic methods, such as radiography, computed tomography (CT), positron emission tomography (PET), histopathology, and next-generation sequencing, are often time-consuming, costly, or require expert interpretation. These challenges are especially pronounced in resource-constrained countries like India, where widespread access to panel-based next-generation sequencing is limited. In addition, issues related to tissue adequacy in lung core biopsies and inherent intratumoral heterogeneity further emphasize the need for an AI-driven solution capable of autonomously detecting and learning nuanced lung nodule features from PET-CT and histopathological images while predicting the likelihood of mutations in critical oncogenes like EGFR.

This talk aims to showcase the true potency of partnership and the boundless potential of AI in advancing healthcare. You will hear about the remarkable journey of partnership with BPGbio & the US Department of Defense and the Department of Energy, to produce groundbreaking advancements in the field of cancer diagnosis using AI. For nearly nine years, BPGbio has collaborated closely with the DoD, leveraging BPGbio's vast biobank of clinically annotated samples and domain-specific AI models, as well as Frontier, the world’s fastest supercomputer through an exclusive partnership with the DoE to discover breakthrough therapeutics and diagnostics. This innovative approach has allowed for the development of cutting-edge diagnostic and screening tests for various cancers, including the development of a prostate cancer screening test that unveiled the groundbreaking biomarker, filamin-A, from years of military samples. The partnership also produced a breast cancer panel targeting ER-positive patients who did not respond to hormone therapy, leading to the discovery of 34 genes that may provide more insight into the cancer's metastatic potential than pathology. The successful outcome of this collaboration has demonstrated the transformative impact of the biology-first AI approach.

Making multi-omic data actionable in a clinical setting requires the integration and processing of a variety of data sources and the availability of a semantically-linked, oncology-focused knowledge graph. This talk provides an overview of practical solutions that address the challenges and complexity involved with interpreting biomarkers, identifying patient cohorts, and finding potential treatment options.