Can blood reveal the body? Trustworthy AI for tissue gene expression (SHERIFF_E27DTP)

A blood sample can be collected repeatedly, whereas tissues such as brain, liver or intestine are much harder to access. Could molecular signals in blood reveal what is happening in other tissues—and tell us when that inference is trustworthy?

This PhD will develop Blood2Tissue, a tissue-conditioned AI framework addressing a fundamental biological challenge: inferring tissue-specific gene-expression programmes from blood. Existing studies demonstrate feasibility for some genes and tissues, but predictions are often population average, accuracy varies substantially, and current approaches rarely identify which individual predictions can be trusted.
You will build on CHROMATIN, our team’s cross-modal attention framework for imputing missing modalities in multi-omics data, and MOCHA, our uncertainty-quantification framework. Blood2Tissue will explore a hybrid transformer/VAE architecture that learns shared and tissue-specific molecular relationships while providing calibrated confidence for each prediction. Transcriptomics will provide the secure core, with additional molecular modalities incorporated where suitable matched datasets are available.

You will benchmark existing methods using GTEx dataset, develop and calibrate Blood2Tissue, and use IBDverse and reference cell atlases to investigate fundamental questions such as why some molecular programmes are blood-visible while others remain tissue-autonomous.
Training will span transcriptomics, machine learning, uncertainty quantification, genetics, single-cell genomics, HPC, reproducible software and responsible AI, including supervised use of emerging AI-assisted and agentic tools for coding, benchmarking and workflow automation. You will join Dr Rahuman Sheriff’s AI integrated Digital Twins Group within Earlham Institute’s data-intensive bioscience environment, with Prof Irene Papatheodorou contributing expertise in cellular genomics and FAIR data, and undertake research visits to Dr Ellen McDonagh’s Open Targets team at EMBL-EBI. The project will deliver reusable computational methodology and new biological understanding of cross-tissue regulation, with longer-term relevance to Digital Twins of living systems.