Precision oncology has long been driven by the ambition to deliver the right treatment to the right patient. But as our understanding of tumour biology becomes more sophisticated, treatment decisions based on individual biomarkers or broad cancer classifications may no longer be enough.
In a recent Oxford Global Thought Leadership interview, Mohammad Ghosheh, CEO of Nexomic, discussed how artificial intelligence and multi-omic data could help oncology move beyond population averages towards a deeper, more dynamic understanding of individual tumour biology.
Moving Beyond Population-Average Medicine
For Ghosheh, one of the fundamental challenges in oncology is the enormous biological variation that exists between patients — even when they share the same cancer diagnosis.
Tumours evolve, develop resistance and respond differently to treatment. Nexomic is therefore developing what Ghosheh describes as a tumour-agnostic AI foundation model, designed to integrate complex multi-omic information and uncover similarities between cancers that might not be apparent from their anatomical classification alone.
Internal analyses, he explained, have highlighted examples where molecular profiles within one cancer type appear more closely related to tumours originating elsewhere than to other patients carrying the same conventional diagnosis.
This supports a growing shift towards classifying and treating cancers according to their underlying molecular biology rather than simply where they originated.
Connecting the Multi-Omic Picture
Current clinical practice still frequently relies on individual biomarkers such as HER2, KRAS or PD-1-related markers to inform patient stratification and treatment decisions.
These biomarkers remain enormously important, but Ghosheh believes AI offers an opportunity to interpret a much richer biological picture.
Genomic, transcriptomic and proteomic information have historically often been analysed independently. AI-powered approaches can increasingly integrate these different datasets simultaneously, revealing biological relationships and signatures that would be difficult to identify by examining each layer in isolation.
This could enable researchers to better understand tumour behaviour, identify clinically meaningful patient subgroups and predict treatment response more accurately.
Discovery Is Only the Beginning
The challenge, however, is not simply discovering new computational signatures.
As AI and multi-omic platforms proliferate, Ghosheh argues that validation, reproducibility and clinical utility are becoming the critical hurdles.
Rich datasets can generate compelling associations, but biomarkers must demonstrate that they remain robust across different patient populations and datasets while avoiding problems such as model overfitting.
For AI-derived biomarkers to influence drug development or clinical practice, discoveries ultimately need to move through rigorous validation and implementation.
The question is therefore shifting from can we identify a signal? to can that signal reliably support a real clinical decision?
Smarter Trials and Better Patient Stratification
The potential impact extends beyond clinical diagnostics.
Better biological stratification could allow drug developers to identify the patients most likely to respond to a therapy, design more targeted clinical trials and potentially improve the probability of demonstrating therapeutic benefit.
This could become particularly important as AI-enabled drug discovery expands the number of potential therapeutic candidates and biological targets entering development.
The ability to connect those candidates with the most appropriate patient populations may become an increasingly important part of translating innovation into successful medicines.
The Patient Remains the End Point
Over the next five years, Ghosheh expects the convergence of AI, multi-omics and precision medicine to accelerate the shift towards increasingly personalised oncology.
The ultimate goal remains straightforward: delivering the right therapy to the right patient at the right time, informed by a deeper understanding not only of the patient's current tumour biology, but also how that biology may evolve and develop treatment resistance.
If AI-powered biomarker platforms can successfully bridge discovery, validation and clinical implementation, their greatest contribution may not simply be finding more biomarkers. It may be helping transform those biological insights into better treatment decisions — and ultimately better outcomes for patients.
