Koohy Group: Applications of multi-omics and AI to decode T cell recognition code in time and space

About the Research 

An effective T cell response is orchestrated upon T cell recognition of MHC-presented antigens on the surface of infected cells or specialized antigen-presenting cells.

A deeper understanding of the rules underpinning T cells recognition of antigens will have implications across numerous disease contexts: the ability to identify the source of T cell autoreactivity could provide new insights into development and progression of most autoimmune diseases. It would enable accurate prediction of cancer neoantigens as targets for personalized cancer vaccines, or prediction of immune escaping mutations in viral outbreaks.

The research interests in the Koohy’s group are therefore focused on development of machine-learning and statistical model to help us better understand the grammar of adaptive T cell immunity.

Here are a few examples showcasing a few active research projects:

We aim to combine the cutting-edge machine-learning and computational advances with emerging state-of-the-art multimodal single-cell experimental technologies to investigate variation of adaptive immune responses to pathogenic and immune-mediated diseases and/or to vaccination at a cellular and tissue level. We are specifically interested in providing a deeper insight into the rules of antigen-specific T cell response in time and space by exploring the dynamics of the cellular and molecular landscape, in health and over the course of a disease or vaccination. We aim to build a map between TCR and their cognate peptide-MHC (pMHC) complexes and to leverage the power if emerging cutting edge deep neural network models such as reinforcement- and representation-learning to deorphanize orphan TCRs that nowadays are routinely generated by commonly used single-cell platforms such as CITE-Seq.

Emerging cutting-edge single cell technologies have enabled quantification of various molecular readouts (transcriptomics, epigenetics, proteomics and so on) at a single cell level from healthy and diseased individuals at different conditions and/or time points. Extracting biologically relevant knowledge from these high dimensional and sparse ‘big data’ has turn out to be very challenging and almost impossible without development of advanced machine-learning techniques. We therefore will continue modelling cellular cross-talk and composition by development of machine-learning models and computational platforms to facilitate integrative analyses of high-dimensional multimodal single cell data.

Recent advances in artificial intelligence and experimental biology have offered breakthrough solutions to problems such as protein structure prediction that were long thought to be intractable. Despite numerous transformative applications of these deep learning models in various biological contexts, they often lack the ability to generalize to out-of-distribution data. For example, AlphaFold that has revolutionized the field of structural biology by accurate prediction of well-structured single chain proteins, is reported to underperform predicting structural and functional properties of short peptide sequences such as T (and B ) Cell Receptors TCRs (and BCRs). It can not also accurately predict   the  structure of multi-chain proteins such as a TCR bound to an antigen which is presented by an MHC molecule ie, TCR:pMHC,  despite their immunes importance in human health. We aim to improve these issues by integration of domain-specific knowledge into large language models. The project will start by a systematic evaluation of performance of existing deep neural network – that originally developed for predicting multi-chain protein complexes—on predicting structure of TCR and TCR:pMHC complexes. By doing so, we aim to pinpoint the main sources of poor performances of these models. The project will then continue to explore potential solutions and incorporation of domain-specific knowledge. It will ultimately turn into devising new deep neural network models to improve the accuracy of predicting TCR and TCR:pMHC structures

Training Opportunities

Our projects and work are mostly of multidisciplinary nature, and usually lie at the intersection of data science (machine-learning and statical inference), immunology and mathematical modelling and therefore motivated students with any of these backgrounds are encouraged to apply. Each project is supervised by a minimum of two supervisor (one from each discipline) to ensure proper training and supervision. Students will additionally have access to a wide variety of training and courses within Oxford University teaching and training schemes.

Students will be enrolled on the MRC Weatherall Institute of Molecular Medicine DPhil Course, which takes place in the autumn of their first year. Running over several days, this course helps students to develop basic research and presentation skills, as well as introducing them to a wide range of scientific techniques and principles, ensuring that students have the opportunity to build a broad-based understanding of differing research methodologies.

Generic skills training is offered through the Medical Sciences Division's Skills Training Programme. This programme offers a comprehensive range of courses covering many important areas of researcher development: knowledge and intellectual abilities, personal effectiveness, research governance and organisation, and engagement, influence, and impact. Students are actively encouraged to take advantage of the training opportunities available to them.

As well as the specific training detailed above, students will have access to a wide range of seminars and training opportunities through the many research institutes and centres based in Oxford.

The Department has a successful mentoring scheme, open to graduate students, which provides an additional possible channel for personal and professional development outside the regular supervisory framework. We hold an Athena SWAN Silver Award in recognition of our efforts to build a happy and rewarding environment where all staff and students are supported to achieve their full potential.

Additional supervisors

Publications: