Spatial mapping of molecular processes in the living brain through integration of transcriptomic, imaging and genomic data
Code: BBSRC-DFA_2026_14
Primary Supervisor: Dr Zhongbo Chen
Email: zhongbo.chen@qmul.ac.uk
Institute: Centre for Preventive Neurology
Secondary Supervisor: Prof Greg Slabaugh
Email: g.slabaugh@qmul.ac.uk
Institute: Digital Environment Research Institute
Abstract:
Understanding how genetic variation relates to spatially organised molecular architecture in the human brain remains a central challenge in spatial biology. This doctoral project aligns with the BBSRC’s strategic priorities in data-driven bioscience and advanced AI by developing novel machine learning frameworks to integrate multimodal spatial datasets and decode molecular organisation from non-invasive imaging.
We hypothesise that quantitative MRI contains latent, spatially structured molecular information that can be decoded using artificial intelligence models trained on spatial transcriptomics. Leveraging 10x Visium spatial transcriptomic profiles from 96 human donors alongside matched single-nucleus RNA-sequencing data, the candidate will develop cross-modal registration algorithms and scale-aware deep learning models to align imaging-derived microstructural features with cell-type proportions and pathway-level gene expression signatures. Emphasis will be placed on interpretable AI, enabling biological attribution of imaging signals to specific cellular processes.
These models will then be deployed in the UK Biobank imaging cohort (>50,000 participants) to generate in vivo predictions of spatial molecular phenotypes at population scale. Finally, the project will test whether inherited genetic variation shapes predicted spatial molecular architecture, establishing imaging-derived molecular representations as an intermediate spatial phenotype linking genotype and tissue-level organisation.
By bridging transcriptomics, imaging, and genomics through advanced multimodal representation learning, this project will develop scalable AI frameworks for cross-scale spatial data integration and contribute foundational methodology data-intensive bioscience.
Lay Summary:
Understanding what is happening inside the living human brain is one of the biggest challenges in neuroscience. Unlike blood or skin, brain tissue cannot be safely sampled from living patients, limiting our ability to directly measure gene activity and cellular organisation within the brain.
This project aims to overcome that barrier using advanced artificial intelligence tools (AI). We will combine detailed molecular maps of genes which are switched on and off in certain parts of the brain (spatial transcriptomics) generated from donated human brains after death with high-resolution brain scans (MRI) and genetic data collected from living study participants.
Spatial transcriptomics, a powerful new technology, allows us to see which genes are active in specific locations within the brain at a very fine level. By aligning these molecular maps with MRI images, we can generate AI models to recognise patterns that link brain structure and function to underlying gene activity.
Once trained, these computational models will analyse MRI scans and look for hidden patterns that link genetic information from living patients to predict the molecular processes likely occurring in their brains—without needing direct tissue samples. We aim to use brain imaging combined with spatial transcriptomics as a “window” into brain biology.
By developing scalable AI tools for integrating transcriptomics, imaging, and genomics, this project will advance computational methods for multimodal spatial biology and provide a general framework for studying tissue organisation when direct sampling is not possible.
Aims and Objectives:
Aim 1: Develop scale-aware AI frameworks for cross-modal alignment of quantitative MRI and spatial transcriptomics
Aim 2: Evaluate scalability and generalisability of AI-derived spatial molecular representations in large population imaging cohorts
Aim 3: Model genetic modulation of spatial molecular representations