Data Before Models: GSK Expands AI Drug Discovery Collaboration with $110 Million Agreement

AI drug discovery models can process the full breadth of existing data, but they may be constrained by the quality of the data itself. GSK and UK biotechnology company Relation Therapeutics have expanded their collaboration, signing a research agreement worth a total of $110 million. They plan to actively generate experimental data in cellular systems relevant to human disease to understand disease mechanisms and identify potential therapeutic targets.

The two companies will build a large-scale human cell perturbation dataset to observe how molecular activity changes over time after cells undergo genetic or pharmacological interventions. Relation plans to generate high-resolution multi-omics readouts through an automated laboratory; the company claims that the facility can produce data at petascale with a level of consistency that conventional laboratories would find difficult to achieve.

The data will be used to train models including MORGAN. MORGAN stands for “Multi-Omic Regulatory Genomics with Artificial Neural Networks” and is a foundation model for cellular perturbation that Relation launched on the same day the agreement was announced. It is not designed solely for a single disease. Instead, it is intended to learn how different cell types respond to genetic and drug interventions, then infer biological mechanisms that may affect disease.

Relation CEO David Roblin said the collaboration will combine physiologically relevant human disease systems, automated experiments, and computational methods. This design addresses a fundamental issue in biomedical AI: publicly available data accumulated across different studies often comes from varying conditions and may not demonstrate causal relationships. By altering cells one at a time in a standardized environment and measuring the consequences, models may obtain leads that are more consistent and easier to validate in the laboratory.

However, cellular responses are not equivalent to treatment outcomes in whole organs or patients. Currently available public information does not disclose named targets, drug candidates, independent validation results for the model’s predictions, the payment structure of the $110 million agreement, or the research and development timeline. Whether MORGAN can be applied across cell types and diseases must still be demonstrated using data not involved in training, replicated experiments, and subsequent drug development results.

Background

This is not the first collaboration between Relation and GSK. In 2024, the two companies signed an agreement focused on fibrotic diseases and osteoarthritis, under which Relation would build functional disease datasets and identify new targets; the new collaboration expands data generation and general-purpose model capabilities to a larger scale. The expanded experimental capabilities are also expected to support Relation’s own programs in fibrotic diseases, osteoarthritis, and osteoporosis, with the research results then fed back to refine MORGAN. If this cycle of repeated calibration between experiments and models proves effective, its value will lie not merely in a larger AI model, but in more reliable disease biology. Until named targets and preclinical results emerge, it remains an investment in a platform rather than a therapeutic breakthrough.