Enhancing Personalized Chemotherapy for Ovarian Cancer: Integrating Gene Expression Data with Machine Learning

Ovarian cancer's complexity and heterogeneity pose significant challenges in treatment, often resulting in suboptimal chemotherapy outcomes. This study aimed to leverage machine learning algorithms, gene selection, and gene expression data to improve chemotherapy results.

The mutual_info_classif approach was employed to identify the most informative genes for predicting treatment responses. Ten machine learning techniques were used to assess and optimize the predictive potential of these genes.

By examining the reciprocal relationships between gene expression and chemotherapy outcomes, the study identified a subset of 20 critical genes essential for treatment efficacy. Among the selected genes, the Random Forest classifier demonstrated the highest accuracy, achieving 97% accuracy, 98% precision, 97% recall, and a 97.5% F1-score in predicting treatment responses. With statistical significance (p = 0.019), the carboplatin predictor successfully distinguished between platinum-sensitive and platinum-resistant patients. Additionally, the combined predictor for the platinum-taxane regimen revealed a significant difference in survival between predicted responders and non-responders, with median survival times of 12.9 months and 8.1 months, respectively (p < 0.045).

The exceptional performance of this model highlights its ability to integrate complex gene expression data, facilitating the development of personalized chemotherapy regimens.