Text mining of CHO bioprocess bibliome: Topic modeling and document classification

Area of Science:

• Biotechnology
• Bioinformatics
• Pharmaceutical Sciences

Background:

• Chinese hamster ovary (CHO) cells are critical for therapeutic protein production.
• Optimizing CHO cell line performance is a growing area of research.
• Bibliographic analysis is needed to understand trends and gaps in CHO bioprocess literature.

Purpose of the Study:

• To qualitatively and quantitatively analyze CHO bioprocess literature using topic modeling.
• To compare Latent Dirichlet Allocation (LDA) model-generated topics with human-curated labels.
• To develop supervised models for identifying new CHO bioprocessing research papers.

Main Methods:

• Manual compilation of a CHO bioprocess bibliome in 2016.
• Application of Latent Dirichlet Allocation (LDA) for topic modeling.
• Development and evaluation of Logistic Regression models for document classification using three CHO bibliome datasets (Bioprocessing, Glycosylation, Phenotype).

Main Results:

• Significant overlap was found between LDA-generated topics and manually assigned categories.
• Machine-generated topics exhibited distinct characteristics.
• Supervised models demonstrated explainability through the use of top terms as features for classifying new papers.

Conclusions:

• Topic modeling effectively captures the landscape of CHO bioprocess research.
• Supervised classification models can accurately identify relevant new publications in the field.
• This approach aids in understanding research trends and identifying future directions in CHO cell line development.