Matrix factorization for biomedical link prediction and scRNA-seq data imputation: an empirical survey

Area of Science:

• Biomedical data analysis
• Bioinformatics
• Computational biology

Background:

• High-throughput technologies generate massive biomedical data.
• Biomedical link prediction and scRNA-seq data imputation are vital for understanding complex diseases.
• Both tasks can be framed as matrix completion problems, often addressed by matrix factorization.

Purpose of the Study:

• To comprehensively review matrix factorization methods for biomedical link prediction and scRNA-seq data imputation.
• To empirically compare representative matrix factorization methods on real-world biomedical datasets.
• To provide guidelines for selecting appropriate methods and identify future research avenues.

Main Methods:

• Systematic literature review of matrix factorization techniques.
• Selection of representative matrix factorization methods.
• Empirical comparison of selected methods on 15 diverse biomedical datasets.
• Performance evaluation across various scenarios.

Main Results:

• Identified challenges in applying matrix factorization to sparse, high-dimensional biomedical data.
• Demonstrated the effectiveness of various matrix factorization methods in biomedical link prediction and scRNA-seq data imputation.
• Established performance benchmarks for different methods across multiple datasets.

Conclusions:

• Matrix factorization is a powerful approach for biomedical matrix completion tasks.
• Guidelines are provided for method selection based on empirical evidence.
• Future research should focus on improving matrix factorization for enhanced biomedical data analysis.