MSclassifier: median-supplement model-based classification tool for automated knowledge discovery.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-throughput technologies generate vast biomedical datasets, necessitating efficient computational tools for knowledge discovery and medical decision support.
• Existing computational models require enhancement to effectively exploit these large-scale datasets for accurate classification tasks.

Purpose of the Study:

• To introduce MSclassifier, a novel machine learning tool designed for automated and effective binary classification.
• To improve classification model performance by incorporating a median-supplement approach for data balancing.

Main Methods:

• MSclassifier utilizes a median-supplement approach, estimating feature medians to generate supplementary data for training sets.
• The tool was evaluated on binary classification tasks, including HER2 receptor expression status in breast cancer and protein subcellular localization prediction.
• Performance was assessed using independent sample and cross-validation tests, comparing MSclassifier against established tools.

Main Results:

• MSclassifier achieved a statistically significant higher classification rate for HER2 status identification (90.30%) compared to the best existing tool (89.83%, p=8.62e-3).
• For protein subcellular localization, MSclassifier demonstrated high performance (93.42%), comparable to another leading tool and outperforming Naive Bayes classifiers.
• The tool proved user-friendly and R-portable, accessible to both programmers and non-programmers.

Conclusions:

• MSclassifier offers a robust and effective method for binary classification in biomedical research.
• Its median-supplement approach enhances model performance, providing a valuable tool for data analysis and decision-making.
• The package's accessibility broadens its applicability across various binary classification challenges in science.