Subtype Characterization of Ovarian Cancer Cell Lines Using Machine Learning and Network Analysis: A Pilot Study

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Ovarian cancer is a complex disease with molecular subtypes impacting patient outcomes.
• High-dimensional mRNA data offers insights but faces challenges in dimensionality and noise.
• Existing classification methods may not fully capture subtype-specific transcriptional patterns.

Purpose of the Study:

• To develop a computational strategy for robust ovarian cancer subtype characterization.
• To reduce dimensionality and identify key molecular features from transcriptomic data.
• To enable better understanding of subtype-specific biology and therapeutic targets.

Main Methods:

• A multi-stage feature selection framework was designed for high-dimensional mRNA data.
• Unsupervised filtering (variance-based, correlation pruning) and supervised methods (Select-K Best, RFE, random forests, LASSO) were employed.
• Gene co-expression similarity networks were constructed using selected discriminative transcripts.

Main Results:

• The pipeline reduced ~65,000 gene features to 83 discriminative transcripts.
• Four distinct ovarian cancer groups were identified based on molecular profiles.
• These groups correlated with known subtypes, TP53 mutations, homologous recombination deficiency, PI3K/AKT signaling, and drug resistance patterns.

Conclusions:

• Combining unsupervised and supervised feature selection with network modeling is effective for ovarian cancer stratification.
• This approach enables robust identification of subtype-specific biological features.
• The findings support improved classification and potential therapeutic strategies for ovarian cancer subtypes.