Emerging Era of Computational Precision Biomarkers

Computational pathology is moving rapidly from promise to practice, reshaping how precision biomarkers are discovered, validated, and deployed in oncology trials.

In a recent virtual thought leadership panel moderated by Dr. Darren Davis, Senior Vice President, Global Digital Pathology, Genomics & Liquid Biopsy Solutions (AI/ML) at Precision for Medicine, experts from PathAI and Roche Tissue Diagnostics discussed how AI-enabled pathology is changing biomarker development, companion diagnostics, and regulatory strategy, particularly in antibody-drug conjugate and bispecific therapy programmes.

The panel featured Eric Walk, Chief Medical Officer at PathAI; Christoph Majewski, Lifecycle Leader PHCS, Personalized Healthcare Solutions at Roche Tissue Diagnostics; and Jennifer Tucker, Pathology Regulatory Network Lead at Roche Tissue Diagnostics. Jennifer Tucker will also be speaking at Biomarkers, CDx & Precision Medicine Basel, where many of these themes around companion diagnostics, regulatory readiness, and precision medicine implementation will continue to be explored.

A central theme of the discussion was reproducibility. Manual pathologist scoring remains a major source of variability in target prevalence and epidemiology studies, often making it difficult to distinguish true biological differences from differences between laboratories. Computational pathology offers a powerful solution by applying consistent scoring across sites, studies, and patient populations. Even when algorithms are not perfect, their reproducibility can reduce confounding and improve confidence in biomarker interpretation.

The panellists also highlighted the growing importance of multimodal datasets. Combining computational pathology with clinical, molecular, IHC, RNA, and sequencing data may allow sponsors to identify more predictive biomarker signatures than any single modality can provide alone. While multimodal approaches have long been discussed, the field is now beginning to generate evidence that they can outperform established methods in certain settings.

For regulators, the key question is whether an AI-driven pathology score is analytically robust and clinically meaningful. Tucker emphasised that strong validation, early clinical evidence, and clear justification of algorithmic cutoffs will be essential before these tools can guide patient selection in pivotal trials. Walk added that regulators are increasingly receptive to models in which the pathologist acts as a quality-control checkpoint rather than the primary scorer, particularly when sponsors engage early through emerging software and breakthrough device pathways.

Another major shift is the move from categorical to continuous biomarker scoring. Traditional pathology thresholds often reflect what humans can reliably score rather than how biology behaves. Computational tools can evaluate biomarker expression as a continuum and identify clinically relevant cutoffs based on outcome data. However, the panellists cautioned that this creates a need for rigorous independent validation to avoid overfitting when large numbers of tissue-derived features are explored.

Global regulation remains complex. The United States, Europe, and China are shaping many of the frameworks that other jurisdictions may follow. In Europe, IVDR has raised expectations for oversight of diagnostic assays used in clinical trials, even before their predictive value is fully established. Sponsors running multinational studies will need proactive engagement, harmonised strategies, and careful planning across regions.

Looking ahead, the panel identified multiplexing, spatial analysis, and even H&E-based companion diagnostics as major opportunities. Computational pathology could also offer commercial advantages by reducing site-specific training once algorithms are locked and enabling faster rollout across clinical networks. At the same time, implementation challenges—including cybersecurity, interoperability, and consistency across multicentre trials—must be addressed.

The discussion closed with reflections on Roche’s announced acquisition of PathAI. Walk described the move as a natural evolution of a partnership that has progressed from algorithm development to companion diagnostic collaboration and now full integration. Majewski pointed to the value of continuous support across the development spectrum, while Tucker highlighted the opportunity to combine PathAI’s strengths in areas such as predetermined change control plans and interoperability with Roche’s established companion diagnostics infrastructure.

The message from the panel was clear: computational pathology is no longer a future concept. It is becoming a practical foundation for the next generation of precision biomarkers, with the potential to improve trial design, regulatory confidence, diagnostic consistency, and ultimately patient selection in oncology.