Why biological data matters more in AI drug discovery

Details

GSK has signed a new research collaboration with Relation Therapeutics valued at up to $110 million, expanding earlier partnerships focused on fibrotic diseases and osteoarthritis. This time the focus is on generating large-scale biological datasets — specifically measuring cellular responses to genetic perturbations and drug interventions — and using those datasets to train AI models within Relation's MORGAN platform to identify drug targets. The partnership reflects a growing recognition in biopharma that data quality, not model sophistication, is the binding constraint in AI drug discovery. Relation's Lab-in-the-Loop approach combines laboratory experimentation with computational analysis: tissue profiling, single-cell transcriptomics, perturbation experiments, and machine learning work in a closed loop. The challenge the collaboration addresses is real: public biological data repositories like CZ CELLxGENE (100M+ standardized cells) and Human Cell Atlas provide training data, but cross-study technical artifacts, sampling protocol differences, and dataset overlap create noise. Proprietary, purpose-built datasets generated in controlled conditions are increasingly where biopharma AI companies differentiate.