A new artificial intelligence framework developed by Indian scientists promises to transform cancer diagnosis and treatment by analysing the disease at a molecular level rather than relying solely on conventional staging.
Researchers from the S N Bose National Centre for Basic Sciences, an autonomous institute under the Department of Science and Technology (DST), in collaboration with Ashoka University, have created OncoMark—an AI model capable of reading the hidden biological programs that drive cancer progression.
Cancer is governed by “hallmarks” such as metastasis, immune evasion and genomic instability, which traditional systems like TNM staging often fail to capture. OncoMark identifies these hallmark activities by analysing massive datasets, offering a deeper understanding of how and why tumors behave differently even when classified under the same stage.
The team led by Dr. Shubhasis Haldar and Dr. Debayan Gupta trained the model on 3.1 million single cells from 14 cancer types, generating synthetic “pseudo-biopsies” that reveal hallmark-driven tumor states. The system achieved over 99% accuracy during internal testing and maintained above 96% accuracy across several independent patient cohorts.
OncoMark was further validated using 20,000 real-world patient samples from eight major datasets, demonstrating its broad clinical potential. For the first time, researchers were able to visually map how hallmark activity intensifies as cancer advances.
Published in Communications Biology (Nature Publishing Group), the study highlights how the tool can guide personalised therapy by identifying which hallmark pathways are active in a patient’s tumor. This could help clinicians choose targeted drugs more effectively and flag aggressive cancers that may appear less dangerous through routine staging.