ST-LDAW: A Topic-Model and Damped Weighted Least-Squares Method for Integrative Deconvolution of Single-Cell and...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Integrating single-cell RNA sequencing (scRNA-seq) with spatial transcriptomics (ST) maps cell types to tissue architecture.
• Existing methods struggle with high-dimensional gene space, leading to noise sensitivity and biased rare cell type detection.

Purpose of the Study:

• To develop a robust computational framework, ST-LDAW, for accurate cell type deconvolution in spatial transcriptomics data.
• To improve the reliability of cell type proportion inference, especially for rare or low-abundance cell types.

Main Methods:

• ST-LDAW employs probabilistic topic modeling to reduce dimensionality and mitigate gene noise.
• Damped weighted least squares optimization enhances robustness by constraining unstable features and preventing overfitting.
• The framework integrates representation and inference levels for improved deconvolution.

Main Results:

• Benchmarking on simulated data showed ST-LDAW achieved 94% recall and 80% accuracy, outperforming existing methods.
• ST-LDAW reliably identifies cell types in complex, sparse datasets, demonstrating robust performance with rare cell types.
• Application to breast cancer data revealed subtype-specific cellular composition and intercellular communication.

Conclusions:

• ST-LDAW offers a significant advancement in spatial transcriptomics analysis, providing more accurate and reliable cell type deconvolution.
• The framework's robustness makes it suitable for complex biological tissues and sensitive detection of rare cell populations.
• ST-LDAW facilitates deeper insights into tissue heterogeneity, communication, and disease mechanisms.