HL-BscPF: Hybrid learning facilitates brain cell auto-identification in multiple pathologies.

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Single-cell transcriptomic data in brain research is rapidly growing in scale and complexity.
• Traditional methods struggle to efficiently extract meaningful insights from this data.
• There is a critical need for advanced computational approaches, particularly artificial intelligence.

Purpose of the Study:

• To introduce the Hybrid Learning-based Brain single-cell Prediction Framework (HL-BscPF).
• To automate cell type classification in brain single-cell data.
• To identify disease-related pathways in the brain.

Main Methods:

• HL-BscPF integrates ItClust and TOSICA models.
• It employs autoencoder-based dimensionality reduction and transformer architecture.
• The framework was evaluated on diverse brain scRNA-seq datasets and benchmarked against ground-truth annotations.

Main Results:

• HL-BscPF accurately classified cell types across aging, Alzheimer's disease, postoperative cognitive dysfunction, and stroke datasets.
• It uncovered key functional alterations in neuronal and glial populations.
• TOSICA excelled in large datasets, while ItClust was optimal for lower cell diversity, showcasing complementary strengths.

Conclusions:

• HL-BscPF demonstrates high accuracy and interpretability in cell type annotation and functional analysis.
• The framework uncovers critical disease-related mechanisms in brain pathologies.
• HL-BscPF serves as a powerful tool for advancing single-cell research in neuroscience.