Identifying Ovarian Cancer-Associated EV mRNA Expression Profiles Using Unsupervised Machine Learning and...

Area of Science:

• Biotechnology
• Bioinformatics
• Molecular Biology

Background:

• Extracellular vesicle (EV) transcriptomics offer insights into cellular states but are complex due to noise and small sample sizes.
• Existing supervised methods for EV data analysis can be biased and struggle with uncovering latent structures.
• Interpreting high-dimensional EV mRNA profiles requires robust analytical approaches.

Purpose of the Study:

• To develop an unsupervised machine learning framework for analyzing extracellular vesicle (EV) transcriptomic data.
• To identify latent gene expression programs and interpretable features from noisy, small-scale datasets.
• To provide a robust method for feature prioritization and representation learning in high-dimensional biological data.

Main Methods:

• A structured preprocessing pipeline involving expression filtering, variance selection, ANOVA, and correlation pruning was implemented.
• Non-negative matrix factorization (NMF) was employed to decompose EV mRNA profiles into gene modules and sample-specific patterns.
• Model selection utilized reconstruction error and component stability; feature prioritization integrated module loadings, group differences, and stability.

Main Results:

• The unsupervised NMF framework successfully extracted structured and interpretable signals from small-scale EV transcriptomic datasets.
• A stable low-rank representation capturing dominant data patterns was identified.
• A compact set of informative features was effectively prioritized using a composite ranking score.

Conclusions:

• Unsupervised matrix factorization provides an effective approach for analyzing challenging EV transcriptomic data.
• The proposed framework enhances feature prioritization and representation learning for high-dimensional biological datasets.
• This method offers a robust alternative to supervised analyses, particularly for small sample sizes.