How Visium helped a top pharma company identify visual biomarkers to improve cancer drug discovery

Challenge: Accelerating biomarker discovery

Finding discriminative biomarkers is essential for efficient clinical operations. They guide better clinical trial design, enable precise patient recruitment, and improve the likelihood of treatment success. Our client, a global pharmaceutical company, sought to identify new visual biomarkers on digital pathology slides of tumor tissues from early-stage clinical trials to distinguish responders from non-responders before advancing to later phases.
Detecting these biomarkers manually, however, is a tedious and often unfeasible process. It requires extensive collaboration between pathologists, data scientists, and clinicians, and even then, valuable image data often remains underused due to time and resource constraints.

How we helped

We applied interpretable ML techniques to large-scale pathology datasets from early clinical trials. By analyzing high-resolution digital slides, the model identified candidate biomarkers, patterns, and features that distinguish responders from non-responders. We then used statistical analysis to refine these findings, narrowing them down to a set of relevant and significant features. The final step involved expert pathologist validation, ensuring that only the most promising biomarkers were retained for clinical use.
This hybrid approach, combining AI discovery power with expert medical validation, enabled the client to extract value from complex image data and accelerate translational research.
‍

We are currently conducting further investigations on larger datasets and the success of this approach could significantly speed up the process of clinical trials, and therefore reduce the cost of clinical development and speed up the time-to-market for our client's drug development projects.

The impact: Unlocking new insights for precision medicine

Integrating AI with human pathology expertise enabled the discovery of previously overlooked biomarkers, leveraging computational scale and analytical depth to uncover biological patterns invisible to traditional human analysis.

• We identified 3 potential new biomarker candidates, fostering fostering new scientific discussions among pathologists.
• Identifying biomarker candidates from images helped leverage an otherwise underutilized data asset (H&E images
• As our approach proved to be promising, we are currently conducting further investigations on larger datasets.