Biotech · 1 min read

The Data Moat: Relation's $110M GSK Pact and the MORGAN Cellular Foundation Model

Relation Therapeutics expands GSK partnership to $110M, unveils MORGAN cellular foundation model, and commits to an automated London wet lab generating petascale perturbation data for AI drug discovery.

By Classy AI News · July 30, 2026

The Data Moat: Relation's $110M GSK Pact and the MORGAN Cellular Foundation Model

Drug discovery AI keeps winning headlines for models. On July 30, 2026, Relation Therapeutics made the harder claim: the competitive edge is owning the wet lab that feeds them.

Relation announced an expanded GSK collaboration worth up to $110 million and unveiled MORGAN — Multi-Omic Regulatory Genomics using Artificial Neural Networks — a cellular foundation model trained on perturbation data Relation generates itself.

Yellow and white balloons on orange surface abstract composition

Causal biology at scale

Under the deal, Relation will build a new automated wet lab near its London headquarters. Cells will be exposed to drugs and genetic perturbations around the clock; multi-omic readouts feed model training at a consistency CEO David Roblin said conventional labs cannot match.

Roblin described the loop to Endpoints News as causal biology: "You make a perturbation and you measure an impact… done at massive scale."

MORGAN: tissue-specific cellular foundation models

MORGAN is designed as a general-purpose cellular model specialized to tissue types and disease areas. Roblin told Endpoints: "We are going to see transformationally different success rates." Relation retains model ownership; GSK receives insights to prosecute into medicines.

Black and red light abstract illustration

Building on a December 2024 pact

The original GSK collaboration targeted fibrosis and osteoarthritis genetics. The July 30 extension deepens infrastructure spending — Roblin confirmed the new lab is funded under the deal.

GSK joins partners including Novartis and Deerfield Management.

The data moat thesis

Public biological datasets are fragmented. Relation's pitch is that proprietary perturbation data at petascale beats better algorithms on stale inputs.

Large display of blue lights in a dark room

Newsletter

Get the dispatch

One field. One email when we publish. Privacy.