AI Unlocks Cancer’s Secrets For Personalized Therapy

A new scientific breakthrough by Indian researchers from SN Bose National Centre for Basic Sciences may reshape the future of cancer treatment.

The researchers introduced an artificial intelligence (AI) framework called OncoMark that decodes cancer not merely as a mass of abnormal cells, but as a living system driven by internal biological motives.

The new model shifts the way doctors look at cancer—from size and spread to the hidden molecular programs that make the disease aggressive, resistant and recurrent.

A scientist from the Department of Science and Technology (DST) explained that cancer has long been studied as a disorder of uncontrolled cell division, yet it is far more complex.

Behind every tumor lies a network of behaviours known as the hallmarks of cancer—mechanisms that allow malignant cells to grow, invade, mutate, escape immune attack, develop new blood supply, and survive treatment.

Two patients may share the same tumour size or stage, yet respond very differently to therapy because the “molecular personality” of their cancer is not the same. Traditional staging systems like TNM cannot fully explain such variation, often leaving oncologists without precise insight into what drives tumor behaviour.

This is where the new study marks a turning point.

The scientists at the SN Bose National Centre for Basic Sciences, an autonomous institute under the DST, in collaboration with Ashoka University, developed this first AI model capable of reading the molecular signature of cancer with remarkable accuracy.

The work was led by Dr. Shubhasis Haldar and Dr. Debayan Gupta, who combined computational biology and AI to map disease at a single-cell level.

The researchers analysed 3.1 million single cells across 14 different cancer types, generating what they call synthetic pseudo-biopsies—computational tumour simulations built from hallmark patterns. Instead of simply identifying the presence of cancer, OncoMark evaluates how key hallmarks interact.

For instance, if immune evasion is high but metastasis is low, the treatment plan may differ from a cancer showing aggressive migration and genomic instability.

In internal validation, the AI model achieved over 99% accuracy, and performed impressively across external datasets—maintaining above 96% accuracy when tested on 20,000 real patient samples from major global repositories.

Importantly, it could visually trace how hallmark activity intensifies from early to late stages of disease, offering clinicians a dynamic picture of tumour progression.

Such capability could transform patient care. The framework can help identify cancers that may appear manageable under standard staging but are genomically primed to progress rapidly. Early recognition means early intervention—a factor known to significantly improve survival outcomes.

It also helps doctors understand which hallmarks are most active in a specific patient, enabling targeted therapy selection. Drugs could be chosen not only to shrink a tumour, but to disable the exact biological mechanism keeping it alive.

Beyond clinical use, OncoMark opens new research possibilities. It may predict which patients are more likely to relapse after treatment, respond poorly to chemotherapy, or develop resistance to existing drugs—challenges that continue to burden oncologists worldwide.

With AI mapping these patterns, new drug discovery could accelerate, guiding pharmaceutical scientists to vulnerabilities in tumour biology previously hidden to the human eye, said the scientist.

The findings are published in Communications Biology (Nature Publishing Group).