GSK and Relation Expand AI Drug Discovery Collaboration
Relation Therapeutics announced on July 30 that it expanded its GSK research collaboration, with up to $110 million in upfront and success-based milestone payments. Relation will generate time-resolved human cellular perturbation datasets with multi-omics readouts and use them to train foundation models, including MORGAN, for therapeutic target discovery; no model-performance or drug-candidate results were disclosed.
Relation Therapeutics announced on July 30 that it expanded its research collaboration with GSK to generate human cellular perturbation datasets and train AI foundation models for drug-target discovery. Relation may receive up to $110 million in upfront and success-based milestone payments; the announcement does not disclose how much is guaranteed upfront.
The agreement extends a relationship established in 2024 around fibrotic diseases and osteoarthritis. Relation's official announcement says the new work will measure how human cells respond to genetic and pharmacological interventions and use the resulting data to develop, validate, and train models, including its MORGAN platform.
Data generation sits beside model training
Perturbation data records biological measurements before and after an intervention, such as a gene edit or drug treatment. Relation says integrated automation will produce time-resolved data with multi-omics readouts at scale. Multi-omics datasets combine measurements across biological layers rather than representing a cell through a single assay.
Fierce Biotech independently reported that Relation is eligible for the $110 million through upfront and success-based milestone payments. It also noted that the companies did not disclose specific disease targets for the expanded collaboration.
MORGAN stands for Multi-Omic Regulatory Genomics using Artificial Neural Networks. Relation describes it as a cellular foundation-model platform intended to predict responses to genetic and pharmacological perturbations across disease contexts. That description is a company claim about the platform's design; the retrieved sources do not report benchmark results, validated therapeutic targets, or drug candidates produced by the new collaboration.
Why the dataset is central
The collaboration places experimental data generation alongside model development rather than treating foundation-model training as a standalone software exercise. BioXconomy reported that Relation plans to use automated laboratories to generate high-resolution multi-omic perturbation datasets and that the same data-production capacity will support both the GSK work and Relation's internal programs.
For ML practitioners in biology, the structure reflects a recurring constraint in scientific modeling: usefulness depends on the coverage, reproducibility, and experimental design of the training corpus. Time-resolved intervention data can be valuable because it links molecular measurements to an explicit change, although predictive performance still requires validation in relevant biological systems.
The companies have not disclosed the new collaboration's disease targets, payment split, experimental scale, model metrics, or timelines for validated discoveries. The agreement is therefore evidence of investment in proprietary data and model infrastructure, not evidence that MORGAN has produced clinically useful targets or shortened drug-development timelines.
The practical signal is the operating model: automated wet-lab generation, multi-omic measurement, and machine learning are being contracted together. Whether that combination improves target selection will depend on results that are not yet public.
Key Points
- 1Relation may receive up to $110 million in upfront and success-based milestone payments; the guaranteed upfront amount was not disclosed.
- 2The collaboration couples automated, time-resolved human cellular perturbation experiments with multi-omic data generation and foundation-model training.
- 3The retrieved sources report no model benchmarks, validated therapeutic targets, drug candidates, or development-timeline improvements from the expanded collaboration.
- 4For scientific ML teams, the material development is the joint investment in proprietary experimental data and model infrastructure, not a demonstrated clinical outcome.
Scoring Rationale
The up-to-$110-million collaboration is a notable deployment of foundation-model methods in pharmaceutical target discovery, with specialized biological data generation as a core deliverable. It is relevant to scientific ML practitioners, while the absence of model metrics, validated targets, drug candidates, and a disclosed payment split limits claims about demonstrated impact.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
