Data-Driven Cell Engineering of Chinese Hamster Ovary Cells through Machine Learning

Jose Manuel Camacho Zaragoza

Research output: Book/ReportPh.D. thesis

397 Downloads (Pure)

Abstract

Over the past decades, advances in bioprocess engineering of mammalian cell cultures have achieved significant improvements in the productivity of recombinant proteins in CHO cells. In contrast, improvements through cell line engineering have remained more modest, possibly due to the complex interwinding of cellular pathways leading to protein secretion. The availability of massive amounts of transcriptomic data and recent developments in computational methods such as machine learning algorithms represent new avenues to explore the complexity of the molecular mechanisms underlying phenotypes such as high protein secretion and guide engineering efforts to fully exploit the biosynthetic potential of mammalian cells.

In this work, we trained a neural network with thousands of gene expression profiles from human cell types of varying secretion capacities and used this model representation to identify common cellular and
metabolic traits that would provide professional secretory cells their characteristic secretion capacity. Functional analysis of the predicted candidate genes highlights key biological functions such as translation, protein metabolism, protein processing, and protein trafficking as essential pathways for enhanced secretion. We further implemented an experimental approach to generate stable pools of producer cell lines to test the effect of induced overexpression of a panel of selected gene candidates on specific productivity. Results showed varying levels of product-specific improvements. Finally, attempts at multiplexed activation of gene candidates using Cas12a and crRNA arrays are presented.

This work is based on two main assumptions:

1) Specialized secretory cells share common cellular and metabolic traits that enable them to secrete
proteins at high rates.

2) These traits are reflected in the transcriptome and can potentially be uncovered by computational
methods.

We hypothesized that these secretory traits might be transferable to CHO cells to improve the production of biopharmaceuticals.
Original languageEnglish
PublisherTechnical University of Denmark
Number of pages101
Publication statusPublished - 2022

Fingerprint

Dive into the research topics of 'Data-Driven Cell Engineering of Chinese Hamster Ovary Cells through Machine Learning'. Together they form a unique fingerprint.

Cite this