Artificial intelligence is beginning to reshape what is possible in biomarker discovery, real-world evidence generation, and precision medicine. In a recent Oxford Global thought leadership interview, Hoifung Poon, General Manager at Microsoft Research, discussed how modern AI can help researchers and clinicians make better use of complex, multimodal patient data.
Poon explained that traditional biomarkers often focus on relatively narrow signals, such as individual genomic variations. While valuable, these may not fully capture the heterogeneity between patients. AI offers a way to integrate high-dimensional data from medical imaging, multi-omics, historical health records, comorbidities, and clinical context into a more holistic patient representation.
A major opportunity lies in structuring unstructured medical data. Clinical notes, pathology reports, imaging text, and genomic reports contain rich information about a patient’s journey from diagnosis through treatment and outcomes. Traditionally, abstracting this data has required significant manual effort, time, and cost. According to Poon, frontier AI could help structure this information in seconds and at far lower cost, potentially accelerating real-world evidence generation and clinical trial operations.
AI may also transform patient matching for precision medicine trials. Clinical trial eligibility criteria can be lengthy and complex, while relevant patient information is often buried in unstructured notes. Poon highlighted the potential for AI systems to continuously process clinical data, identify eligible patient cohorts, support feasibility analysis, and improve trial recruitment.
Looking ahead, Poon sees AI’s impact across a spectrum: from productivity gains, such as automating tedious data abstraction, to creativity gains, where AI helps uncover new biomarker patterns that humans could not easily detect. One long-term vision is the development of “digital twins” or virtual patient representations that can help forecast disease progression and treatment response.
For biomarker discovery and precision health, success over the next five years may depend on moving beyond simple biomarkers toward more sophisticated, multimodal, AI-enabled approaches. While the field is still some distance from its most ambitious goals, Poon suggested that AI could play a critical role in helping researchers learn from real-world patient journeys at a scale no human expert could achieve alone.
