Geometry of Senescence: Spatial AI for Population- and Perturbation-Dependent Heterogeneity
Code: BBSRC-DFA_2026_10
Primary Supervisor: Dimitrios Kollias
Email: d.kollias@qmul.ac.uk
Institute: School of Electronic Engineering and Computer Science
Secondary Supervisor: Cleo Bishop
Email: c.l.bishop@qmul.ac.uk
Institute: Blizard Institute
Abstract:
Cellular senescence is a key hallmark of ageing and a contributor to cancer, fibrosis and chronic
inflammatory disease. The targeted clearance of senescent cells using senolytics offers the promise
of improved healthspan. However, senescent cells are highly heterogeneous: they differ in
morphology, molecular programmes, spatial organisation and response to drugs. Current analytical
approaches typically treat this variability as a clustering problem, aiming to define discrete subtypes.
This project proposes a conceptual and methodological shift: senescent heterogeneity will be
modelled as structured geometry within morphology-molecular-space, shaped by spatial context,
perturbation and population variables.
Using large-scale single-cell spatial datasets combining high-resolution imaging with multiplex RNA
and protein phenotyping, we will develop a spatially-regularised multimodal manifold framework.
Each cell will be embedded in a shared latent space that integrates morphology, molecular profiles
and neighbourhood structure. Rather than counting clusters, heterogeneity will be quantified through
geometric metrics including dispersion, intrinsic dimensionality and curvature.
Drug treatments, genetic manipulations and demographic groups (age, sex and ethnicity) will be
modelled as probability distributions over this manifold. Optimal transport-based methods will quantify
distributional shifts, allowing us to distinguish variance expansion, bifurcation and spatial niche
formation. Cross-modal modelling will determine which molecular layers best explain morphological
geometry and identify regions of molecular-morphological concordance or ambiguity.
The project will deliver novel spatial AI methods for modelling heterogeneity as geometry rather than
taxonomy, alongside open-source tools for image-based spatial biology. This framework will advance
understanding of senescent cell organisation and provide generalisable methodology for analysing
complex intra and inter-model heterogeneity.
Lay Summary:
As we age, some of our cells enter a state called senescence. These cells stop dividing but remain biologically active, releasing signals that can contribute to inflammation and disease. Importantly, senescent cells are not all the same. They vary in shape, behaviour, molecular activity and how they are organised within tissues. Understanding this variation is essential for developing effective therapies that target harmful senescent cells while preserving beneficial functions.
Most current approaches group cells into categories or “subtypes”. However, this can oversimplify the complexity of ageing tissues. This project proposes a new way of thinking about senescence. Instead of defining fixed subtypes, we will treat variation as a continuous landscape. Using advanced AI, we will analyse high-resolution spatial images of individual cells together with detailed molecular and morphological measurements and their positions within tissues.
The AI system will learn how cells differ in appearance and molecular state and how these differences are organised in space. It will then measure how this landscape changes after drug treatment and whether these changes differ between younger and older individuals or between males and females.
By developing new computational tools that model cellular variation as a structured landscape rather than fixed categories, this project aims to provide deeper insight into how ageing tissues are organised. The methods developed will also be applicable to other diseases, including cancer and fibrosis, where cellular diversity plays a critical role.
Aims and Objectives:
Aim 1: Learn how senescent cells look and quantify heterogeneity geometry.
Aim 2: Explain why cells cluster after drug perturbation.
Aim 3: Map morphology to molecular explanation.
Aim 4: Model population-aware heterogeneity.