CL-Impute: A contrastive learning-based imputation for dropout single-cell RNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed cell heterogeneity studies.
• Dropout events in scRNA-seq data compromise downstream analysis reliability.
• Current imputation methods struggle with inaccurate cell relationships derived from noisy data.

Purpose of the Study:

• To develop a novel imputation method for scRNA-seq data that addresses challenges posed by dropout events.
• To improve the accuracy and reliability of gene expression imputation in scRNA-seq datasets.
• To overcome limitations of existing methods that rely on potentially untrustworthy cell relationships.

Main Methods:

• Proposed CL-Impute (Contrastive Learning-based Impute) model.
• Utilized contrastive learning to learn cell representations from dropout events.
• Employed a self-attention network to capture global cell relationships.

Main Results:

• CL-Impute demonstrated superior performance over state-of-the-art methods in quantitative assessments, cell clustering, gene identification, and trajectory inference.
• The combination of contrastive learning and masking cell augmentation enabled learning of actual latent features from noisy data.
• Enhanced reliability of imputed values in scRNA-seq data with high dropout rates.

Conclusions:

• CL-Impute is an effective contrastive learning-based method for imputing scRNA-seq data, particularly in high dropout scenarios.
• The method learns robust cell representations by focusing on dropout events and global cell interactions.
• Source code is available for the CL-Impute model, facilitating its adoption and further research.