scMFF: a machine learning framework with multiple feature fusion strategies for cell type identification.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate cell type classification is essential for single-cell RNA sequencing (scRNA-seq) data analysis.
• Existing methods often rely on single feature types, failing to capture the full complexity of cell type differences.
• Naïve feature concatenation can introduce noise and redundancy, hindering model performance.

Purpose of the Study:

• To develop an advanced framework for robust cell type classification in scRNA-seq data.
• To address the limitations of single-feature representations in capturing cell type heterogeneity.
• To improve the performance and stability of scRNA-seq data analysis through effective feature fusion.

Main Methods:

• Proposed scMFF, a multiple feature fusion framework integrating four distinct feature types.
• Explored six different fusion strategies to optimize feature integration.
• Evaluated various classifiers in combination with the proposed fusion methods.

Main Results:

• scMFF demonstrated superior performance compared to single-feature approaches across 42 disease-related datasets.
• The framework showed enhanced stability in cell type classification.
• Validation on an external COVID-19 dataset confirmed the robustness and effectiveness of scMFF.

Conclusions:

• Multiple feature fusion offers a more comprehensive and reliable approach to scRNA-seq cell type classification.
• scMFF provides a stable and effective solution for analyzing complex single-cell data.
• The proposed framework advances the field of computational biology for disease-related research.