Data4Cure presented two new research abstracts at the 2025 American Association for Cancer Research (AACR) Annual Meeting, highlighting AI-based approaches for drug-target discovery and large-scale cancer subtyping using foundation models applied to large collections of sequencing data.
Knowledge-graph AI-based prioritization of drug target candidates
In the first study (Abstract LB112/13), Data4Cure researchers leveraged the company's large-scale biomedical knowledge graph encompassing over 4 billion relations across more than 1 million entities — including genes, diseases, drugs, pathways, cell types, and tissues. To prioritize candidate cancer drug targets, they trained and optimized 17 state-of-the-art knowledge-graph embedding (KGE) and graph neural network (GNN) models. The best models significantly outperformed standard machine-learning approaches, uncovered clusters of predicted targets enriched for specific biological functions, and identified novel candidates — including within previously underexplored functional classes.
Foundation-model integration of >180,000 bulk RNA-seq samples
The second study (Abstract LB343/6) describes RNA1, a novel transformer-based foundation model trained on over 180,000 bulk RNA-seq profiles. RNA1 enables systematic assignment of cancer samples to molecular subtypes and the discovery of new subtypes across more than 36,000 tumor samples. RNA1-based subtyping shows significant associations with survival and drug-response outcomes on published clinical-trial datasets, outperforming previously published subtyping approaches across a majority of evaluated benchmarks.
Both projects build on Data4Cure's Biomedical Intelligence Cloud platform and its mission to help turn complex multi-modal data into knowledge that drives biomedical research and therapeutic innovation.