Personalising Multiscale Models with Spatial Transcriptomics to Investigate T Cell Exclusion Mediated by Cancer-Associated Fibroblasts in Non-Small Cell Lung Cancer
This repository contains code and resources developed during a research project at Institut Curie that aimed to personalise multiscale models (MSMs) of tumour-immune interactions using spatial transcriptomic data. The focus was on understanding mechanisms of T cell exclusion mediated by cancer-associated fibroblasts (CAFs) in non-small cell lung cancer (NSCLC), particularly lung squamous cell carcinoma (LUSC).
Spatial organisation within the tumour microenvironment (TME) significantly influences immune cell infiltration. CAFs are known to form barriers around tumour nests, excluding T cells and contributing to resistance to immunotherapy. This project explored CAF-mediated T cell exclusion using in silico multiscale simulations combining agent-based and Boolean intracellular models.
The MSMs were personalised using spatial transcriptomic (ST) data from LUSC samples, deconvoluted to infer cell types and gene expression at single-cell resolution.
Multiscale Modelling (MSM): Built using the PhysiBoSS framework, integrating:
- Agent-based modelling with PhysiCell
- Boolean modelling with MaBoSS
Spatial Transcriptomic Data:
- Visium v2 data from NSCLC patient samples (not avaialable on GitHub)
- Deconvolution using Cell2location and SpatialScope
Model Personalisation Pipeline:
- Cell localisation and annotation via BioInformatics WalkThrough (BIWT)
- data_analysis: Scripts for exploratory analysis and preprocessing of spatial and single-cell transcriptomic data.
- cell2location: Scripts to create the environment, run Cell2location, and analyze the results.
- spatialscope: Scripts to create the environment, run SpatialScope, and analyze the results.
- models: Aberrant cell cycle Boolean model from Sizel et al., 2019.
- personalisation: Deconvolution results transformed into initial conditions for the agent-based model, including initial cell populations derived using the transformed results and implemented with PhysiCell and the Bioinformatic WalkThrough tool.
To reproduce or run the personalised simulations:
Set up the environment: Use the provided Apptainer/Singularity definition files to ensure reproducibility across systems.
Preprocess ST and scRNAseq data and Run deconvolution
Personalise the model: Initialise agent-based models with inferred cell types and spatial locations
Run simulations: Launch PhysiBoSS simulations from the model configurations
Cell2location and SpatialScope deconvolutions matched known histological structures, validated against immunohistochemistry (IHC). Personalised MSMs reproduced distinct tumour architectures and immune exclusion phenotypes. Simulations explored how CAF-derived ECM can mechanically hinder T cell infiltration.
Key references include:
- Grout et al., Cancer Discovery 2022 – CAF states in NSCLC
- Sizek et al., Cell Cycle Model 2019 – Boolean modelling
- Kleshchevnikov et al., Nature Biotechnology 2022 - Cell2location
- Weigert et al., IEEE Xplore 2022 - StarDist
- Wan et al., Nature Communication 2023 - SpatialScope
- Sobkowicz A., Internship Report 2025
This work was developed by Agathe Sobkowicz during her Master’s internship at Institut Curie, under the supervision of Dr. Vincent Noël, Dr. Laurence Calzone, and Pr. Emmanuel Barillot. Please open an issue or submit a pull request if you wish to contribute or have questions.