Artificial Intelligence (AI) is at the cusp of transforming every industry in profound ways by creating a seamless partnership between man and machine. Professor Chalapathy Neti is particularly excited by the potential of AI to transform biomedical research and healthcare. After a long stint at IBM in developing AI technologies and applying it to healthcare and education, Professor Neti is convinced that it can profoundly impact and fast track our journey towards precision medicine. His desire and ambition are to help expedite the development of a “scalable” federated AI platform for early detection and intervention of diseases like cancer, anchored on multi-omics/multi-scale data – an exemplary implementation of the “Cancer Moonshot”. While several efforts are underway, they are still piecemeal, and fraught with many challenges at the technical, social and policy level.

Deploying AI at scale is a challenge and requires significant innovation in federated data architectures that ensure data privacy, security, and intellectual property needs. In this talk, Professor Neti will describe a systematic approach to design and build a federated AI and data architecture on Hybrid Cloud to enable AI at scale and tackle grand challenges like the “Cancer Moonshot”.

For the AI layer, Professor Neti will describe a hierarchical deep learning approach that combines disease progression signatures at multiple scales (molecular, cell-level, tissue-level, organ-level, body-level, and population-level data) for early detection and intervention of cancer progression. For the data layer, he will describe the concept of a comprehensive cancer atlas i.e. a rich multimedia database spanning structured data (clinical attributes/symptoms), text (genetic signatures, clinical notes), and medical images (cytopathology and histopathology of tumors, CT images, etc.). Several efforts are underway to create such atlases (e.g. TCGA, NCI Cancer research Data Commons; International Cancer Proteogenomic consortium, etc.) to advance research.

Professor Neti will discuss the daunting challenges of creating such cancer atlases for federated AI. Data that underpins the care progression for any individual typically spans several years and is distributed across many healthcare institutions (primary, secondary, and tertiary) and departmental silos within an institution (clinical, radiology, pathology, etc.). He will outline the design of the cancer atlas using a federated data architecture on hybrid cloud, to enable a dynamic, longitudinal, secure and privacy preserving (w/ patient in control) architecture. If the cancer atlas is organized appropriately using FAIR (Findable, Accessible, Interpretable, Reusable) principles, linked at multiple levels, like concepts in Wikipedia, and zoomable like “Google maps” (from body-level to molecular-level, similar to satellite to individual houses in Google maps), it will enable and unleash unprecedented use of AI/Machine learning research for disease progression modeling, accurate clinical trial matching for biomarker based clinical trials, and other use cases.