Performance analysis of data resampling on class imbalance and classification techniques on multi-omics data for...

Area of Science:

• Computational biology
• Genomics
• Machine learning in oncology

Background:

• Cancer remains a major global health challenge.
• Early detection and accurate classification are crucial for effective treatment and improved patient outcomes.
• Multi-omics data integration offers a powerful approach for understanding cancer complexity.

Purpose of the Study:

• To develop and evaluate computational models for classifying normal versus tumor samples.
• To integrate RNA sequencing, copy number variation (CNV), and DNA methylation data for enhanced cancer classification.
• To compare various machine learning algorithms and techniques for addressing class imbalance in cancer datasets.

Main Methods:

• Utilized The Cancer Genome Atlas (TCGA) dataset for liver cancer, breast cancer, and colon adenocarcinoma.
• Developed joint analysis models integrating RNA seq, CNV, and DNA methylation data.
• Evaluated 18 machine learning methods using AUC, precision, recall, and F-measure, and compared five class imbalance techniques, with Synthetic Minority Oversampling Technique (SMOTE) showing superior performance.

Main Results:

• The model using Stochastic Gradient Descent (SGD) with Support Vector Machine (SVM) achieved over 99% accuracy and AUC >= 0.999 for liver and breast cancer.
• For colon adenocarcinoma, both SGD and Sequential Minimal Optimization (SMO) achieved 100% accuracy and 1.000 for AUC, precision, recall, and F-measure.
• SMOTE was identified as the most effective technique for handling class imbalance in the cancer datasets.

Conclusions:

• Integrated multi-omics data and machine learning provide highly accurate methods for cancer sample classification.
• SGD and SMO are effective algorithms for tumor classification, demonstrating excellent performance across different cancer types.
• Advanced computational approaches are vital for improving cancer detection and aiding in the fight against cancer.