A classification framework applied to cancer gene expression profiles.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression profiling is crucial for understanding cancer biology and guiding treatment strategies.
• Developing robust machine learning (ML) models is essential for accurate cancer classification, distinguishing between subtypes or normal versus cancerous tissues.

Purpose of the Study:

• To explore and evaluate supervised machine learning techniques for cancer classification.
• To identify optimal gene sets for differentiating cancer subtypes or normal versus cancer samples using a novel feature selection approach.

Main Methods:

• A two-step feature selection process combining the ReliefF attribute estimation method and a genetic algorithm was utilized.
• Supervised learning algorithms including decision trees, k-nearest neighbors (KNN), support vector machines (SVM), bagging, and random forests were applied.
• Five distinct cancer datasets were used for evaluating the performance of various classification methods.

Main Results:

• No single classification method demonstrated universal superiority across all datasets.
• K-nearest neighbors (KNN) and linear support vector machines (SVM) generally exhibited improved classification performance.
• Integrating diverse genomic data, such as protein-protein interaction networks alongside gene expression, significantly increased prediction accuracy compared to using gene expression alone.

Conclusions:

• Machine learning, particularly KNN and linear SVM, shows promise for accurate cancer classification based on gene expression.
• A hybrid feature selection approach effectively identifies informative gene sets for cancer subtyping.
• Multi-omic data integration represents a powerful strategy for enhancing the predictive accuracy of cancer diagnostic models.