GSK's $110M check to Relation is underwriting a human-cell data factory

GSK's $110M check to Relation Therapeutics is underwriting a human-cell data factory. The AI foundation model trained on that data is the second-order asset, and the July 30, 2026 expansion is the third such deal Relation has signed with a major drugmaker in 18 months.

The expanded collaboration sends up to $110M in upfront and milestone payments to the London-based, Nvidia-backed AI biotech FierceBiotech. In return, Relation will run large-scale experiments on human cells, capturing how those cells respond to genetic and drug perturbations, then feed the resulting datasets into its foundation model MORGAN, short for Multi-Omic Regulatory Genomics using Artificial Neural Networks, to suggest new drug targets GlobeNewswire.

GSK and Relation signed two earlier deals in 2024, one in fibrotic diseases (conditions where tissue scars and hardens) and one in osteoarthritis, that established the same data-to-model pipeline TechTimes. The July 30 announcement is that template applied to a still-undisclosed set of disease targets, and the undisclosed targets are themselves part of the story.

Relation's December 2025 pact with Novartis showed how the template travels between pharmas. That deal brought $55M in upfront cash and equity plus research funding, with up to $1.7B in potential milestones, and it focused on atopic diseases, the allergic-inflammation family that includes eczema and certain forms of asthma. The MORGAN model retrained on Relation's atopic data is what Novartis is paying for, and the same model architecture is what GSK is now underwriting in a second disease area. Two pharmas, two disease focuses, one shared data-to-model pipeline.

The asset class is the data, not the model. A foundation model in this context is a large AI system trained on broad biological data and then adapted to predict how cells respond to perturbations across diseases. Once the model exists, retraining it on a new disease area's experimental data is incremental work. Building the experimental data, the perturbation measurements, the cell lines, the controls, the replication, is what costs time and money. That is the moat Relation is selling. CEO David Roblin described the new deal as generating "novel datasets from physiologically relevant human disease systems" to support future discovery GlobeNewswire, which is the way a contract manufacturer would describe a new production line.

The headline figure is also a ceiling. Up to $110M is the maximum, not the check, and the mix tilts heavily toward success-based milestones that pay only if targets from the collaboration advance. The actual committed cash is undisclosed and almost certainly a fraction of the total. That structure is standard for early-stage target identification, but it means the deal is a hedge: GSK is buying optionality on a data pipeline, not funding a clinical program. Against GSK's overall R&D budget, $110M is rounding.

Two claims fall out of the data-factory frame and can be tested. If pharma's internal human-cell data generation scales, or if foundation models trained on publicly available perturbation data become good enough to rival private datasets, the data-template premium collapses. The undisclosed disease targets under the new GSK deal make that test hard to run in the near term. Relation's published pipeline still leans on immunology, metabolic disease, and bone disease, but the company has not said which of those, or which new ones, the MORGAN retraining will cover. Until that changes, the $110M ceiling reads as a vote of confidence in a template rather than a verdict on any specific target.

The next data point to watch is whether a third major pharma signs on. Novartis and GSK now give Relation two anchor partners on opposite sides of the atopic-and-fibrotic map. A third name, in oncology or neurodegeneration, would suggest the template generalizes. A quiet 12 months would suggest the opposite, that the data-factory premium is real but narrow, and that Big Pharma is willing to pay it only one or two bets at a time.