Multimodal foundation models for cellular phenotyping and biological discovery

This PhD project addresses a major challenge in computational biology and biomedical AI: the lack of generalisable models that can learn biologically meaningful representations across diverse data modalities, including microscopy images, tissue-level imaging, and molecular or sequence-based data. While recent foundation models have shown strong performance in computer vision and language, their ability to capture cellular morphology, protein localisation, cellular organisation, perturbation response, and host–pathogen interactions across biological settings remains limited.

The student will investigate and develop multimodal foundation models for cellular phenotyping and biological discovery, with a particular focus on integrating image-based and molecular views of cellular state. The project will explore how self-supervised and weakly supervised learning approaches can be used to align cell-level and tissue-level imaging data with complementary biological information such as protein sequence, genomic data, or whole-genome sequence data. The goal is to build robust and transferable representations that support biological discovery across a range of applications, including protein localisation, perturbation biology, infection biology, and comparative host–pathogen analysis.

Potential research directions include multimodal representation learning for fluorescence microscopy and tissue imaging, image–sequence alignment, biologically informed fusion strategies, and methods for improving generalisation across datasets, cell types, organisms, and experimental conditions. The work may also investigate interpretable and uncertainty-aware approaches to ensure that learned representations are biologically grounded and useful for downstream tasks such as phenotype prediction, mechanism-of-action analysis, infection identification, and cellular response modelling.

The project is supported by an interdisciplinary supervisory team spanning multimodal AI, foundation models, infection biology, and biosciences. The student will be primarily supervised by Dr Syed Sameed Husain, with established collaborations in AI and biosciences, and co-supervised by Dr Lorraine McElhinney at APHA, Defra and Professor Roberto La Ragione in Surrey FHMS, providing a strong interdisciplinary environment and access to expertise in computational modelling, infection biology, and translational bioscience.

Applicants should hold, or expect to obtain, a minimum of a UK 2:1 honour degree (or international equivalent) in Computer Science, Artificial Intelligence, Data Science, Bioinformatics, Computational Biology, Electronic Engineering, or a closely related discipline. A masters degree in a relevant subject is desirable but not essential.

We are seeking a highly motivated candidate with strong technical foundations and an interest in interdisciplinary research at the interface of AI and biosciences. Experience in one or more of the following areas would be advantageous: machine learning and deep learning, multimodal learning, foundation models, computer vision, computational biology, or biological image analysis.

Applicants should demonstrate strong programming skills, ideally in Python, and familiarity with machine learning frameworks such as PyTorch. They should also have good mathematical foundations, particularly in linear algebra, probability, statistics, and optimisation, together with the ability to work independently and communicate research clearly.

Prior research output would be an advantage, ideally evidenced by publications in leading AI/ML or medical AI venues such as NeurIPS, ICML, ICLR, CVPR, MICCAI, or equivalent. Prior experience with microscopy imaging, sequence data, multimodal biological datasets, or large-scale model training would also be beneficial, but is not required.

Open to candidates who pay UK/home rate fees. See UKCISA for further information.