Heuristic principal component analysis based unsupervised feature extraction and its application to bioinformatics Y-h. Taguchi, Dept. Phys., Chuo Univ. We have developed Principal Component Analysis Based unsupervised Feature Extraction Advantages 1. Unsupervised → no need to decide selection criterion, which is decided semi- a priori 2. Computationally “un”challenging
→ Just execute PCA once.
Procedures: 1. Suppose that you have N samples × M features 2. Embed features by PCA 2. Investigate contributions from samples to each PC 3. Select M' (<M)outliers along the specified PC 4. replace N×M with N×M'. 5. Analyze data (e.g., discrimination, biomarker, drug candidate etc)
Achievements: 1. Successful identifications of genes with aberrant promoter methylation among three autoimmune diseases. 2. Proposal of universal disease biomarker that consists of unique 12 circulating microRNAs and can diagnose more than 20 diseases 3. Integrated analysis of mRNA expression and promoter methylation for non-small cell lung cancer. Pl ease visit our poster #7 0!