Speaker
Description
Immunotherapy has varied results and how to predict whether a patient will respond is an open question. It is crucial to closely monitor a patient’s response once immunotherapy begins, and to potentially change to an alternate treatment if needed. Using a mathematical model, Creemers et al. argued that a patient’s tumor growth rate and immune cell killing rate are the primary parameters affecting response \cite{Creemers et al.}. These can be difficult to estimate as they vary stochastically over time. We tackle this by training a deep learning model that takes in noisy time series data of a patient’s tumor growth during treatment, and predicts whether or not they will recover during the next five years. We use a CNN-LSTM architecture designed to detect time series transitions far in advance\cite{Bury et al.}. This model uses CNN layers to extract important features and then uses LSTM layers to detect patterns over time. Heterogeneous patients are generated using a mathematical model with noisy, realistic parameters. The deep learning model is trained on these synthetic patients, and then a reserved test dataset is used to investigate how far in advance the model can accurately predict outcomes. We propose that frequent tumor measurements of an immunotherapy patient can be fed to the model to help assess whether or not to continue treatment, thus improving the patient's quality of life and chances of survival.
Bibliography
@article{Creemers et al., title={A tipping point in cancer-immune dynamics leads to divergent immunotherapy responses and hampers biomarker discovery}, volume={9}, ISSN={2051-1426}, url={https://jitc.bmj.com/lookup/doi/10.1136/jitc-2020-002032}, DOI={10.1136/jitc-2020-002032}, abstractNote={ }, number={5}, journal={Journal for ImmunoTherapy of Cancer}, author={Creemers, Jeroen H A and Lesterhuis, W Joost and Mehra, Niven and Gerritsen, Winald R and Figdor, Carl G and De Vries, I Jolanda M and Textor, Johannes}, year={2021}, month=may, pages={e002032}, language={en} }
@article{Bury et al., title={Deep learning for early warning signals of tipping points}, volume={118}, ISSN={0027-8424, 1091-6490}, url={https://pnas.org/doi/full/10.1073/pnas.2106140118}, DOI={10.1073/pnas.2106140118}, number={39}, journal={Proceedings of the National Academy of Sciences}, author={Bury, Thomas M. and Sujith, R. I. and Pavithran, Induja and Scheffer, Marten and Lenton, Timothy M. and Anand, Madhur and Bauch, Chris T.}, year={2021}, month=sept, pages={e2106140118}, language={en} }
