The proposed fellowship uses multimodal cancer data to help predict the treatment response, personalise patient care, and improve survival and quality-of-life post-treatment. By harnessing diverse data sets, the AI models are trained to identify subtle patterns and indicators. This not only aims to customize treatment plans to individual patient needs but also seeks to mitigate the risks associated with radiotherapy, potentially reducing the occurrence of debilitating side effects and improving overall treatment efficacy. The project will focus on clinical, anatomical, and biological patient data. First, the project focuses on employing AI to analyze complex pathology slide images. This research is set to transform tumor diagnostics by providing unprecedented insights into the microenvironment of cancers. This approach aims to uncover new diagnostic markers and patterns, enhancing the accuracy of tumor classification and staging. The fusion of these AI-generated insights with the recognised prognostic feature of pathologists will lead to more precise results. Furthermore, recognising the gap in the application of AI in clinical settings, a part of the research is focused on developing a web-based platform that makes AI tools readily available and user-friendly for clinicians. The platform is envisioned to provide non-specialist healthcare professionals with access to state-of-the-art AI analysis for radiology and histopathology. This means that clinicians can benefit from AI-powered insights in real-time, enhancing their decision-making process in patient care. This service aims to democratize the use of AI in healthcare, making it a standard part of clinical practice and thus accelerating the adoption of AI in medical diagnostics and treatment planning. Collectively, these components of my research will represent a significant leap forward in the application of AI in the realm of cancer care. The project is not just about technological innovation; it's about fundamentally transforming the approach to cancer diagnosis and treatment.

Harnessing the immune system to treat cancer has revolutionised survival outcomes for many patients. Immune checkpoint inhibitor therapies which unleash the brakes from immune cells to kill cancer cells, have become standard of care for many cancer subtypes. The success of existing, emerging and future immunotherapies and their routine use in the NHS is dependent on the appropriate tools, data and technology to rationalise their use and manage their side effects. Nevertheless, almost no biomarkers today can effectively distinguish responders from non-responders, predict toxicity, or guide treatment choices. Moreover, existing datasets lack standardization, suffer from sampling biases, and fail to integrate 'multi-omics' data with clinical information. Our platform, MANIFEST, leverages existing and also novel scalable methodologies to provide deep profiling of each patient receiving immunotherapy and will deliver on multimodal data integration and modelling. We represent a diverse group of UK-wide experts in cancer research and clinical care comprising 6 major NHS trusts, 14 academic institutes and universities, and with strong upfront top-up investment (>£ 12 million) from industry partners (namely IMU Biosciences, Guardant Health, Natera, Roche-imCORE, Roche-Sequencing, M:M Bio and Microbiotica; among others). To demonstrate the utility of the MANIFEST platform, we will deliver exemplar projects encompassing multiple tumour types (melanoma, renal cell carcinoma, bladder cancer and triple negative breast cancer), where prediction of treatment outcomes and toxicities to both standard of care and emerging immunotherapies remains a significant unmet need. Specifically, we have access to pre-existing longitudinal samples of >3,000 patients across 10 reported studies (RAMPART, MITRE, PRISM, EXACT, CAPTURE, PaVeMenT, ALEXANDRA, neoTRIP, ABACUS and ABACUS-2). In parallel, we will utilise existing governance at partner NHS sites for prospective sample collection (blood, stool and tumour) and processing. With a tiered approach, we aim to profile patient and tumour samples at scale (~3,000 patients over 3 years). We will implement standard of care diagnostic workflows for high-volume biomarker discovery (Tier 1). We reserve in-depth profiling, through Tiers 2 and 3 participation, to further characterise tumours including discovery-focused techniques, such as high-dimensional peripheral immune profiling, liquid biopsy (cfDNA, immune methylation profiling), and spatial image-profiling approaches coupled with molecular profiling (WES, bulk&long-read RNAseq, TCR&BCRseq). Finally, for selected patients, we will apply our expertise in Representative Sequencing (RepSeq), a novel tumour sampling methodology which overcomes pervasive sampling bias in solid tumours; and pre-clinical modelling through patient-derived tumour fragments (PDTFs) for drug sensitivity screening. Finally, we will deploy a team of 12 experts in artificial intelligence and machine learning to deliver on multimodal data acquisition and integration in our in-house Trusted Research Environment (TRE). We are also excited to be teaming up with the National Pathology Imaging Co-operative (NPIC) and Genomics England (GEL) to synergise efforts in translating discoveries for patient benefit.