GSK Expands Collaboration With Relation: Training AI on Cellular Perturbation Data to Identify Disease Mechanisms and New Targets

Whether artificial intelligence in drug discovery can truly understand disease often depends not only on model size, but also on the experimental data used to train it. UK biotech company Relation Therapeutics has announced an expansion of its research collaboration with GSK to generate, at scale, data on changes in human cells following genetic or drug interventions, with the aim of identifying disease mechanisms and potential therapeutic targets from dynamic responses.

According to Relation’s announcement, the agreement includes an upfront payment and performance-based milestone payments totaling up to $110 million. The company did not disclose the amount of the upfront payment, the milestone conditions, the duration of the collaboration, or which asset rights GSK may obtain. Relation will use human disease cell models and automated experimental systems to generate time-series cellular perturbation data while simultaneously measuring changes across multiple molecular layers.

Cellular perturbation involves altering cells through gene editing, regulation of gene expression, or compound treatment, and then observing how RNA, proteins, and other molecular signals reorganize over time. Compared with simply comparing healthy and diseased samples, such data may be more likely to reveal causal direction—for example, which downstream responses disappear or shift after a pathway is shut down. However, whether the results can extend beyond the culture dish and reproduce the complex environment within the human body remains a key hurdle for target validation.

The data generated through the collaboration will be used to develop, validate, and train foundation models, including MORGAN, which Relation announced on the same day. Its name is derived from “Multi-Omic Regulatory Genomics with Artificial Neural Networks.” It is designed to predict how human cells respond to genetic and pharmacological interventions across different cell types and disease contexts. Relation said MORGAN will be supported by petabyte-scale multi-omics data generated through automated high-throughput experiments, with models built for specific cell types of therapeutic importance.

The practical value of the collaboration will depend on whether new targets proposed by the models can be reproduced in independent experiments and subsequently lead to drug candidates with selectivity, safety, and efficacy in humans. Currently available information does not specify the targeted diseases, cellular systems, data scale, or quality-control methods. Nor have MORGAN’s benchmark tests, external validation results, or performance relative to existing methods been disclosed. It is therefore not yet possible to determine how many of its predictions can be translated into developable drug programs.

The agreement indicates that major pharmaceutical companies are willing to invest in research infrastructure combining automated wet-lab experiments, multi-omics measurements, and AI modeling. However, the maximum transaction value is not equivalent to value already paid. More informative signals to watch will be whether the collaboration produces experimentally confirmed disease mechanisms, specific targets advanced by GSK, and results that hold in patient-derived samples or models that more closely reflect human physiology.