Evaluating Ovarian Cancer Chemotherapy Response Using Gene Expression Data and Machine Learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Ovarian cancer (OC) is the leading cause of gynecological cancer deaths in the US.
• Serous ovarian cancer (SOC) is the most common subtype.
• Transcriptomics yields vast gene expression data, but identifying clinically relevant genes is challenging.

Purpose of the Study:

• To develop a computational framework for feature selection (FS) in SOC.
• To identify genes associated with chemotherapy response in SOC patients.
• To build predictive models for drug response using identified gene signatures.

Main Methods:

• Applied LASSO and varSelRF feature selection methods to SOC datasets.
• Utilized Gene Expression Omnibus (GEO) data.
• Employed random forest (RF) and support vector machine (SVM) for model evaluation.

Main Results:

• Identified biomarker panels of 9 and 10 genes for platinum-paclitaxel and platinum-only response, respectively.
• Achieved over 90% accuracy in predictive models trained on identified gene signatures.
• Demonstrated the efficacy of FS in reducing biomarker numbers and enhancing biological relevance.

Conclusions:

• Multiple FS methods effectively reduce biomarker complexity and increase biological relevance.
• The proposed framework accurately predicts drug response in cancer treatment.
• This approach enhances the utility of gene expression data for clinical applications in SOC.