New AI framework decodes cancer’s molecular behaviour, opening pathways for personalised therapy
The framework, published in Communications Biology (Nature Publishing Group), allows clinicians to visualise hallmark activity across cancer stages for the first time.
New Delhi, November 27: Researchers at the SN Bose National Centre for Basic Sciences, in collaboration with Ashoka University, have developed a first-of-its-kind artificial intelligence framework that can decipher the molecular behaviour of cancer—offering a major leap toward personalised treatment and more accurate prognosis.
Cancer has long been classified using systems such as the TNM staging model, which evaluates tumor size and spread. But such approaches often fail to capture the deeper biological programs—known as the hallmarks of cancer—that drive how tumors grow, evade immunity, metastasize and resist therapy. These hallmarks explain why patients with the same clinical stage can experience vastly different outcomes.
The new AI framework, OncoMark, reads the “molecular personality” of cancer by decoding the complex signals inside tumor cells. Developed by a team led by Shubhasis Haldar and Debayan Gupta, the model was trained on 3.1 million single cells across 14 cancer types. Using this dataset, the researchers generated synthetic “pseudo-biopsies” that represent hallmark-driven tumor states, enabling the AI to learn how different biological processes fuel cancer progression.
OncoMark demonstrated exceptional performance—achieving more than 99% accuracy in internal tests and maintaining above 96% accuracy across five independent patient cohorts. It was further validated on 20,000 samples from eight major datasets, confirming its robustness and real-world relevance.
The framework, published in Communications Biology (Nature Publishing Group), allows clinicians to visualise hallmark activity across cancer stages for the first time. This could help identify aggressive tumors that appear less threatening under conventional staging and support earlier, more targeted interventions.
Crucially, OncoMark can reveal which hallmarks are active in an individual patient’s tumor, enabling doctors to match treatments to the underlying molecular processes—such as immune evasion, metastasis or genomic instability—rather than relying solely on anatomical staging.
Researchers say the tool has the potential to reshape cancer diagnostics, guide personalized therapy selection, and deepen scientific understanding of how different hallmarks interact to shape tumor evolution.
The study’s findings mark an important step toward integrating AI-driven molecular intelligence into routine oncology, bringing precision medicine closer to reality for millions of patients.
