The Critical Role of Biological Data in AI Drug Discovery: Lessons from GSK’s Latest Partnership

GSK and Relation Therapeutics: Forging a Path with Data-Driven AI

For readers tracking the shift, The landscape of drug discovery is being reshaped by artificial intelligence, but the true power of AI models hinges on one fundamental element: high-quality biological data. A recent significant collaboration between pharmaceutical giant GSK and British biotechnology firm Relation Therapeutics, valued at up to $110 million, underscores this very principle.

Meanwhile, This expanded partnership isn’t just about applying AI; it’s a strategic investment in generating the precise, large-scale biological datasets essential for unlocking the next generation of therapeutic breakthroughs. This substantial investment empowers Relation to generate extensive datasets that map how human cells react to genetic modifications and drug interventions.

These meticulously created datasets are crucial. They will serve as the training ground for advanced AI models, including those within Relation’s proprietary MORGAN platform, to pinpoint novel drug targets with greater accuracy and efficiency. This collaborative model highlights a key trend: the integrated development of both cutting-edge AI models and the bespoke biological data that fuels them.

Unpacking Relation’s “Lab-in-the-Loop” Methodology

In practical terms, Relation Therapeutics employs a unique “Lab-in-the-Loop” approach, seamlessly blending laboratory experimentation with sophisticated computational analysis. This methodology encompasses a wide array of techniques, from tissue profiling and single-cell multi-omics to advanced sequencing and target validation.

Central to their work are perturbation experiments, which meticulously measure the impact of genetic alterations on cellular characteristics linked to disease. The insights gleaned from these experiments are then rigorously analyzed alongside genetic and patient-derived biological data, creating a comprehensive picture for AI interpretation.

The Double-Edged Sword of Public Biological Datasets

For example, While public repositories like CZ CELLxGENE, the Human Cell Atlas, and NCBI Gene Expression Omnibus offer vast quantities of single-cell data, they come with inherent complexities. A 2025 review in Experimental & Molecular Medicine highlighted that these resources provide access to millions of standardized cells, yet their utility for training robust AI foundation models is not without challenges.

Significant technical hurdles arise from combining data produced across diverse studies. Variations in sampling methods, sequencing protocols, experimental procedures, and processing pipelines can introduce inconsistencies. Furthermore, single-cell data often contains technical noise and artifacts, necessitating stringent dataset selection, filtering, and quality control during model training.

That said, Another critical issue is dataset overlap. The review pointed out that identical or highly similar cells might appear across multiple public resources.

This redundancy can disproportionately influence model training and introduce data-leakage risks if training and test datasets aren’t carefully managed. Ultimately, the review concluded that assembling a high-quality, non-redundant dataset is as vital as the model’s architecture itself.

Beyond “More Data”: The Nuance of AI Model Performance

Counterintuitively, simply increasing the volume of biological training data doesn’t always guarantee superior AI model performance. Research published in Nature Methods in June examined single-cell foundation models, finding that many reached performance plateaus even after training on only a fraction of available data.

Interestingly, Unlike large language models, which often benefit from continuous data scaling, the assessed single-cell systems did not consistently demonstrate improved results with ever-larger training sets. The study emphasized the importance of balancing model capacity, dataset size, and computational resources, rather than just expanding them indiscriminately.

This perspective was echoed in a separate 2025 study in Genome Biology, which evaluated models like Geneformer and scGPT. It found that these advanced models didn’t consistently outperform simpler methods and cautioned against assuming that larger, pretrained models automatically yield better biological representations, especially given challenges like batch effects.

The Strategic Shift Towards Specialized Datasets

However, Recognizing these complexities, pharmaceutical companies are increasingly investing in specialized, high-quality datasets tailored to specific diseases. Relation Therapeutics, for instance, has developed “Osteomics,” a proprietary functional single-cell bone atlas. This project combines patient-derived samples with multi-omics data, imaging, genomics, and clinical phenotypes to investigate osteoporosis.

This trend was further highlighted in a 2025 Nature Biotechnology analysis of AI-focused biopharma deals, which identified specialized dataset providers as a key emerging trend. Examples include GSK’s $37.5 million agreement with Ochre Bio for human liver single-cell data and AstraZeneca’s $200 million partnership with Tempus to develop oncology foundation models using extensive patient data.

Meanwhile, These collaborations underscore a fundamental truth: access to sufficient, high-quality data remains a significant bottleneck in AI drug discovery. Agreements are evolving to address this, ranging from direct data licensing and joint development to the creation of entirely new biological datasets, as seen in the GSK-Relation deal.

Conclusion: The Future of AI Drug Discovery is Data-Driven

The partnership between GSK and Relation Therapeutics exemplifies a crucial shift in AI drug discovery. It’s not just about sophisticated algorithms, but about the strategic generation and meticulous curation of biological data. As the field matures, the ability to produce high-fidelity, disease-specific datasets will be paramount for translating AI’s potential into life-changing medicines.

Expert Perspective

A practical read on biological data AI drug discovery starts with data. That is where the earliest effects are likely to show up if this development keeps building.

What happens next will come down to adoption speed, policy response, and execution quality. That combination could make biological data AI drug discovery a meaningful reference point across models.

For decision-makers, the useful lens is not the headline alone but how training changes priorities once organizations have to respond.

Frequently Asked Questions

Why is biological data AI drug discovery important?

GSK and Relation Therapeutics: Forging a Path with Data-Driven AIFor readers tracking the shift, The landscape of drug discovery is being reshaped by artificial intelligence, but the true power of AI models hinges on one fundamental element: high-quality biological data.

What impact could biological data AI drug discovery have?

A recent significant collaboration between pharmaceutical giant GSK and British biotechnology firm Relation Therapeutics, valued at up to $110 million, underscores this very principle.Meanwhile, This expanded partnership isn’t just about applying AI; it’s a strategic investment in generating the precise, large-scale biological datasets essential for unlocking the next generation of therapeutic breakthroughs.

What should readers watch next with biological data AI drug discovery?

This substantial investment empowers Relation to generate extensive datasets that map how human cells react to genetic modifications and drug interventions.These meticulously created datasets are crucial.

How does this relate to data?

It connects because the article frames data as one of the clearest areas where the topic may be felt in practice.

Source: https://www.artificialintelligence-news.com/news/gsk-relation-therapeutics-ai-drug-discovery-biological-data/