New Delhi—Indian researchers have developed an advanced artificial intelligence framework that could transform how cancer is diagnosed, understood and treated. Instead of judging a tumour only by its size or spread, the new system studies the disease through its molecular behaviour — offering a clearer picture of what drives cancer inside the body.
Scientists explained that cancer is driven by several hidden biological programs known as the “hallmarks of cancer.” These hallmarks describe how normal cells become cancerous, how they grow, escape the immune system, invade organs and resist treatment. Traditional staging systems like TNM focus mainly on tumour size, lymph nodes and metastasis, but they often fail to explain why two patients with the same stage may respond very differently to treatment.
Researchers from the SN Bose National Centre for Basic Sciences and Ashoka University have now introduced an AI framework called OncoMark, which reads the molecular signals of cancer to predict how a tumour will behave. According to the Ministry of Science and Technology, it is the first tool capable of mapping cancer’s internal biological programs with such detail.
The team analysed 3.1 million single cells from 941 tumours covering 14 cancer types. They used this data to generate synthetic “pseudo-biopsies” that capture hallmark-driven tumour states. With this massive dataset, the AI learned how key hallmarks — including metastasis, immune evasion and genomic instability — work together to drive tumour growth and treatment resistance.
The study, published in Communications Biology, describes how OncoMark uses neural multi-task learning to estimate hallmark activity directly from gene expression data. This method helps reveal the real biological processes behind cancer progression, something existing diagnostic tools rarely measure.
During testing, OncoMark achieved over 99 percent accuracy internally and maintained more than 96 percent accuracy across five independent patient cohorts. It was further validated on 20,000 patient samples from eight global datasets, proving its wide applicability. For the first time, scientists were able to visualise how hallmark activity increases as cancer advances.
Experts say this AI tool can identify which hallmarks are active in a patient’s tumour, helping doctors choose treatments that target those specific processes. It can also detect dangerous, fast-growing cancers that may appear less severe under traditional staging, enabling earlier and more precise intervention.
Researchers stated that the framework can predict the activity of 10 cancer hallmarks at once and deliver rapid analysis of tumour data, supporting personalised treatment strategies for patients.
