Track: Poster Competition

Abstract

Chimeric Antigen Receptor T-cell therapy involves the genetic modification of T-cells to find and attack cancer cells throughout the body. Establishing critical quality attributes and parameters (CQAs, CPPs) is a crucial task to ensure potency, safety, and consistency of this therapy. To understand these therapies, an experiment evaluating the expansion of T-cells was done. Through a supervised learning approach, multi-omics predictors can be used to better understand T-cell growth and memory behavior. Hence, the purpose of this work is to develop a computational pipeline to molecularly characterize T-cells using multi-omics profiles and extract those predictive features of quality at early stages of manufacturing. Specifically, this pipeline is designed to measure the predictive power of omics variables and to assess model sensitivity when highly correlated predictors are present. The computational tool incorporates a consensus of important variables between Random Forest (RF), Gradient Boosted Trees (GBT), Support Vector Machines (SVM), and Symbolic Regression (SR). These models showed high prediction performance (R2 :75%-95%) and strong feature consensus (e.g. IL2R, IL21, MIF) when modeling CD4 memory fraction. Many other growth and memory responses were investigated to characterize T-cells. Moreover, an approach involving the clustering of highly correlated variables is proposed to mitigate the computational impact of highly correlated predictors in the modeling process. Preliminary results showed that this methodology is able to rank groups of correlated predictors and improve RF performance. Ultimately, understanding these highly correlated features can better define the feature consensus (CQAs, CPPs) across methods. The findings of this work could enable the discovery of new knowledge necessary to achieve scalable biomanufacturing of cell therapies in an automated manner.