Machine learning approach to identify significant genes and classify cancer types from RNA-seq data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer is a major global health burden, causing millions of deaths annually.
• Current cancer identification methods are slow, costly, and require significant resources.
• There is a critical need for faster, more efficient cancer detection and classification techniques.

Purpose of the Study:

• To evaluate machine learning algorithms for cancer type classification using RNA-seq gene expression data.
• To identify statistically significant genes associated with different cancer types.
• To assess the efficiency and accuracy of various machine learning models in cancer genomics.

Main Methods:

• Utilized the PANCAN RNA-seq dataset from the UCI Machine Learning Repository.
• Assessed eight machine learning classifiers: Support Vector Machines, K-Nearest Neighbors, AdaBoost, Random Forest, Decision Tree, Quadratic Discriminant Analysis, Naïve Bayes, and Artificial Neural Networks.
• Validated model performance using a 70/30 train-test split and 5-fold cross-validation.

Main Results:

• The Support Vector Machine model demonstrated the highest classification accuracy, achieving 99.87% under 5-fold cross-validation.
• Identified statistically significant genes through RNA-seq data analysis.
• Compared the performance of eight different machine learning algorithms for cancer classification.

Conclusions:

• Machine learning, particularly Support Vector Machines, shows significant potential for accurate and efficient cancer classification from RNA-seq data.
• This approach can accelerate biomarker discovery and aid in developing personalized cancer diagnostics and treatments.
• The study highlights the utility of computational methods in advancing cancer research and clinical applications.