12–17 Jul 2026
University of Graz
Europe/Vienna timezone

Integrating Mechanistic Modeling and Machine Learning to Predict CAR-T Therapy Outcomes from CD4+/CD8+ Dynamics

16 Jul 2026, 14:20
20m
15.11 - HS (University of Graz)

15.11 - HS

University of Graz

102
Contributed Talk Mathematical Oncology Contributed Talks

Speaker

Saranya Varakunan (University of Waterloo)

Description

Chimeric antigen receptor (CAR) T cell therapy has shown remarkable success in hematological malignancies, yet patient responses remain highly heterogeneous. Clinical evidence suggests that the CD4:CD8 T cell ratio in CAR-T cell treatments influences therapeutic outcomes \cite{Galli2023}, but the quantitative roles of these subsets in shaping treatment dynamics are incompletely understood. Building on a tumor-regulated effector-exhaustion CAR-T cell framework \cite{Kirouac2023}, we develop a mechanistic model that explicitly represents CD4+ helper and CD8+ cytotoxic T cell populations and their interactions with tumor burden \cite{Boulch2021}.

Due to the absence of detailed clinical datasets, we evaluate the model using virtual patient simulations; we show that our model reproduces qualitative clinical trends in CAR-T expansion and treatment response at varying CD4:CD8 ratios, while also revealing substantial inter-patient variability. We then identify key determinants of therapeutic outcome using sensitivity analyses, and assess predictive performance using these determinants (i.e. how measurement errors for important parameters affect model prediction). We show that a simple feedforward neural network improves the model’s predictive decay under noise, highlighting the potential of hybrid mechanistic and machine learning approaches for forecasting CAR-T therapy outcomes.

Bibliography

@article{Galli2023,

author = {Galli, E. and Bellesi, S. and Pansini, I. and Di Cesare, G. and Iacovelli, C. and Malafronte, R. and others},

title = {The CD4/CD8 ratio of infused CD19 CAR-T is a prognostic factor for efficacy and toxicity},

journal = {British Journal of Haematology},

year = {2023},

volume = {203},

number = {4},

pages = {564--570},

doi = {10.1111/bjh.19117}

}

@article{Kirouac2023,

author = {Kirouac, D. C. and Zmurchok, C. and Deyati, A. and Sicherman, J. and Bond, C. and Zandstra, P. W.},

title = {Deconvolution of clinical variance in CAR-T cell pharmacology and response},

journal = {Nature Biotechnology},

year = {2023},

volume = {41},

number = {11},

pages = {1606--1617},

doi = {10.1038/s41587-023-01687-x}

}

@article{Boulch2021,

author = {Boulch, M. and Cazaux, M. and Loe-Mie, Y. and Thibaut, R. and Corre, B. and Lemaitre, F. and Grandjean, C. L. and Garcia, Z. and Bousso, P.},

title = {A cross-talk between CAR T cell subsets and the tumor microenvironment is essential for sustained cytotoxic activity},

journal = {Science Immunology},

year = {2021},

volume = {6},

number = {57},

pages = {eabd4344},

doi = {10.1126/sciimmunol.abd4344}

}

Author

Saranya Varakunan (University of Waterloo)

Co-authors

Melissa Stadt (University of Waterloo) Mohammad Kohandel (University of Waterloo)

Presentation materials

There are no materials yet.