SAD: Sparse-Aware Diffusion Model for Single-Cell Gene Expression Completion.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data is crucial for understanding cellular heterogeneity.
• Current scRNA-seq datasets often have limited gene coverage (10-12k genes) and technical zeros, hindering downstream analysis.
• Foundation models require comprehensive gene expression data for accurate generalization.

Purpose of the Study:

• To address the limitations of sparse scRNA-seq data for foundation models.
• To introduce a novel gene-completion task and framework for scRNA-seq data.
• To enhance the reliability and completeness of scRNA-seq datasets for advanced applications.

Main Methods:

• Developed SAD, a diffusion-based framework for gene completion in scRNA-seq.
• SAD is designed to handle extremely sparse data and high missing rates.
• The framework corrects sparsity-distribution bias and generates missing gene entries.

Main Results:

• SAD significantly outperforms existing methods in gene completion across various metrics.
• The framework demonstrates superior performance in extreme sparsity scenarios (missing rates >80%).
• SAD provides consistent and reliable gene expression inputs (over 30k genes) for foundation models.

Conclusions:

• SAD offers a robust solution for the gene-completion task in scRNA-seq.
• The framework enables the effective reuse of missing scRNA-seq information.
• SAD lays the groundwork for improved precision medicine applications through enhanced data quality.