Inverse Projection Representation and Category Contribution Rate for Robust Tumor Recognition

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Sparse Representation based Classification (SRC) shows promise but faces limitations like insufficient training samples and underutilization of test data.
• Existing SRC methods exhibit instability in representation, impacting classification performance.
• Effective utilization of test samples is crucial for robust classification in high-dimensional data.

Purpose of the Study:

• To develop a stable Inverse Projection Representation based Classification (IPRC) method that overcomes limitations of traditional SRC.
• To enhance tumor recognition from microarray gene expression data by leveraging test samples more effectively.
• To introduce a robust framework for analyzing gene expression data and identifying pathogenicity-related genes.

Main Methods:

• Proposed a novel Inverse Projection Representation (IPR) and analyzed its feasibility and stability.
• Developed a classification criterion, 'category contribution rate', for matching IPR and performing classification.
• Introduced a statistical measure to quantify the stability of representation-based classification methods.
• Implemented a two-stage hybrid gene selection method for identifying informative genes in microarray data.

Main Results:

• The proposed IPRC method demonstrated effective use of test samples, leading to improved classification stability.
• Extensive experiments on six public tumor microarray datasets showed the IPRC technique is competitive with state-of-the-art methods.
• The framework successfully identified informative genes and provided functional analysis of candidate pathogenicity-related genes.

Conclusions:

• IPRC offers a stable and effective approach for representation-based classification, particularly in analyzing gene expression data.
• The method addresses key limitations of SRC, providing a more robust solution for tumor recognition.
• The developed framework facilitates deeper insights into gene expression data and potential disease biomarkers.