Dr. Tapabrata Rohan Chakraborty (FIET, FHEA, SMIEEE) recently joined the UCL Department of Medical Physics and Biomedical Engineering as a full-time Lecturer (Assistant Professor), and brings with him his research group, the Transparent and Reliable AI Lab (TRAIL), which he will build upon further at UCL. He was previously a principal research fellow with UCL Cancer Institute and an honorary member of staff with UCL MPBE. Rohan is a Theme Lead in Frontier AI Assurance with the Alan Turing Institute (UK’s national institute for AI) and an invited expert in Responsible AI with the Global Partnership on AI (GPAI, part of the Organisation for Economic Co-operation and Development, OECD). He is an Editor with Springer Nature Computer Science. On the teaching side, Rohan will be leading a module on Cancer AI as part of UCL’s new Computational Cancer MSc course.
A computer scientist by training, Rohan’s primary research focus is the development of explainable and trustworthy AI models for precision biomedicine and personalised healthcare, particularly for multimodal computational cancer applications. Earlier this year, he led a high-impact computational study that showed the power of diffusion based cross modal generative AI to synthesise cancer transcriptomic signatures from digital histopathology images. This paper published in Nature Communications further demonstrated that when the information from the digital pathology and the synthesised transcriptomics is combined using cross-attention and sent through a transformer-based multimodal predictive AI model, it can accurately perform cancer grading and survival risk estimation.
Multimodal cancer AI has huge potential to transform clinical practice, but for trusted adoption, can the integration of multimodal cancer data maximise population-level accuracy while limiting individual-level uncertainty of predictions simultaneously in a transparent and reliable manner?
Though the above paper is a computational proof-of-concept study using public TCGA data, Rohan is now working with colleagues at the UCL Cancer Institute to clinically validate his AI pipeline on two large UCL-led trials: STAMPEDE (prostate cancer) and OPTIMA (breast cancer). He is also exploring with industrial partner Veracyte for potential commercialisation down the road. Recently, Rohan’s team extended the work towards prediction of transcriptomics/proteomics from cell painting data for phenotypic drug discovery. The paper was accepted for publication through The International Conference on Machine Learning (ICML), a leading outlet in core AI/ML research, and will be presented at ICML 2026 in Korea, July 2026.
Rohan says “Multimodal cancer AI has huge potential to transform clinical practice, but for trusted adoption, can the integration of multimodal cancer data maximise population-level accuracy while limiting individual-level uncertainty of predictions simultaneously in a transparent and reliable manner?”. This becomes especially timely given the recent MHRC recommendations on enhanced safety requirements for AI based medical devices. Rohan says one way of implementing AI assurance is through personalised uncertainty quantification with conformal prediction, as laid out in his perspective article in Nature Machine Intelligence last year, which was selected for UCL REF.
