Publication:
Clustering highly correlated predictors to extract early predictive signatures of CAR-T cell quality

dc.contributor.advisor Torres García, Wandaliz
dc.contributor.author Odeh Couvertier, Valerie
dc.contributor.college College of Engineering en_US
dc.contributor.committee Dávila-Padilla, Saylisse
dc.contributor.committee González Barreto, David
dc.contributor.department Department of Industrial Engineering en_US
dc.contributor.representative Vásquez Urbano, Pedro
dc.date.accessioned 2020-07-28T09:57:34Z
dc.date.available 2020-07-28T09:57:34Z
dc.date.issued 2020-07-03
dc.description.abstract Improving accessibility to innovative cancer immunotherapies such as CAR-T cell therapy is imperative to allow treatment for patients in need. The establishment of critical quality attributes and parameters is crucial for ensuring the potency, safety, and consistency needed to guarantee a successful large-scale manufacturing transition, which in turn will lower costs and increase accessibility. Hence, this work aims to molecularly characterize T cells and to extract predictive features of quality at early stages of its manufacturing using machine learning techniques that mitigate the impact of multicollinearity. To this end, this work proposes a multi-step semi-supervised approach that incorporates consensus clustering of features before model fitting process and a median position metric that allows an unbiased importance ranking of both clusters and variables inside each cluster. Results demonstrated that this approach was able to mitigate the impact of highly correlated omics features and extract putative driving variables. en_US
dc.description.abstract Aumentar la accesibilidad de pacientes a tratamientos contra el cáncer tales como las terapias con células CAR-T es imperativo. Para reducir costos operacionales se deben identificar los atributos y parámetros críticos para la calidad y seguridad de estas terapias que aseguren una transición exitosa a una manufactura a grandes escalas. El propósito de este trabajo es caracterizar molecularmente las células T y extraer características predictivas de calidad en las primeras etapas de su fabricación utilizando técnicas de aprendizaje automatizado que reduzcan el impacto de la alta correlación entre variables. Este trabajo propone un enfoque semi-supervisado por etapas que incluye agrupar variables altamente correlacionadas y una métrica de posición de la mediana que permite una clasificación de la importancia de estas variables. Los resultados demostraron que este enfoque fue capaz de mitigar el impacto de características ómicas altamente correlacionadas y de extraer variables predictivas de la calidad de estas terapias. en_US
dc.description.graduationSemester Summer en_US
dc.description.graduationYear 2020 en_US
dc.description.sponsorship This material is based upon work supported by the National Science Foundation under Grant No. EEC-1648035. en_US
dc.identifier.uri https://hdl.handle.net/20.500.11801/2633
dc.language.iso en en_US
dc.rights.holder (c) 2020 Valerie Odeh Couvertier en_US
dc.subject CAR-T cells en_US
dc.subject High correlation en_US
dc.subject Random Forest en_US
dc.subject CQA en_US
dc.subject VIM en_US
dc.title Clustering highly correlated predictors to extract early predictive signatures of CAR-T cell quality en_US
dc.type Thesis en_US
dspace.entity.type Publication
thesis.degree.discipline Industrial Engineering en_US
thesis.degree.level M.S. en_US
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
ININ_OdehCouvertierV_2020.pdf
Size:
64.17 MB
Format:
Adobe Portable Document Format
Description:
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.88 KB
Format:
Item-specific license agreed upon to submission
Description: