Cancer classification and pathway discovery using non-negative matrix factorization.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Understanding genetic mutations is crucial for disease research.
• Accurate cancer type prediction aids in diagnosis and treatment.

Purpose of the Study:

• To develop a novel method for predicting cancer type using whole-exome sequencing data.
• To aggregate genetic mutations effectively for improved disease classification.

Main Methods:

• Applied non-smooth non-negative matrix factorization (nsNMF) for dimensionality reduction.
• Utilized support vector machine (SVM) for cancer type classification.
• Scored and collapsed somatic mutations from whole-exome sequencing data.

Main Results:

• Achieved an average prediction accuracy of 80% in distinguishing four cancer types.
• The proposed method significantly outperformed baseline models.
• Identified key genes and pathways associated with specific cancer types.

Conclusions:

• The developed pipeline offers a generic approach for studying somatic mutations and cancer.
• The method demonstrates potential for disease status classification and pathway discovery.