Kidney Week 2025 Early Program - Advances in Research Conference 2025: Generative Artificial Intelligence in Kidney Disease Research and Management-Panel Discussion: Will AI Help Find New Disease Mechanisms and Therapeutic Targets?

Video Transcription

Video Summary

The panel discusses how “AI” should be defined in biomedicine, contrasting broad machine learning with today’s focus on generative AI. Speakers agree AI has already helped identify disease mechanisms and therapeutic targets, especially by integrating multiple evidence types (e.g., human genetics for efficacy, single-cell data for safety) to prioritize targets and reduce clinical trial failure. They highlight limits of supervised learning for discovering truly novel mechanisms and emphasize the promise—but also constraints—of unsupervised approaches, which mainly interpolate within known data rather than extrapolate to entirely new biology.<br /><br />A recurring theme is that progress depends less on sheer data volume than on high-quality, standardized, openly shared datasets and infrastructure. Kidney research is catching up to oncology-style resources (e.g., KPMP analogous to TCGA), and open science is presented as essential for accelerating discovery and industry collaboration.<br /><br />Practical opportunities include pathology automation (e.g., counting glomeruli), EHR-based prediction of CKD progression using text/structured data, and multimodal modeling. Panelists caution against forcing omics into LLM text; instead, align modalities via embeddings and databases. They advise pharma to invest in curated data, expertise, compute, and coordinated consortia, while keeping expectations realistic about “one model for everything.”

Asset Subtitle

Patrick Aloy, Casey S. Greene, Julio Saez-Rodriguez, Pinaki Sarder, Jovan Tanevski

Meta Tag

Module ARC

Speaker FACULTY FACULTY

Keywords

AI definition in biomedicine

generative AI vs machine learning

target identification and prioritization

multimodal evidence integration (genetics and single-cell)

unsupervised learning limits in biology discovery

open standardized biomedical datasets and infrastructure

kidney precision medicine (KPMP) and CKD prediction

pathology automation and EHR-based modeling