High-throughput single-cell RNA-seq data imputation and characterization with surrogate-assisted automated deep...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data.
• Gene sparsity in scRNA-seq data presents computational challenges for analysis.
• Designing deep neural network architectures for imputation requires specialized expertise.

Purpose of the Study:

• To develop an automated method for designing deep neural network architectures for scRNA-seq data imputation.
• To improve the accuracy and efficiency of gene expression imputation in scRNA-seq datasets.
• To provide a tool that reduces the need for manual tuning of deep learning models.

Main Methods:

• Surrogate-assisted Evolutionary Deep Imputation Model (SEDIM) was developed for automated architecture search.
• SEDIM utilizes an offline surrogate model to accelerate the search process.
• The model was evaluated on imputation and clustering tasks using scRNA-seq data.

Main Results:

• SEDIM significantly improved imputation and clustering performance compared to benchmark methods.
• The automated architecture design process was validated across different datasets and platforms.
• Exploration in mass cytometry and metabolic profiling demonstrated SEDIM's versatility.

Conclusions:

• SEDIM offers an effective and efficient solution for deep neural network architecture design in scRNA-seq imputation.
• The method provides novel insights into cell-type identification and underlying biological mechanisms.
• SEDIM's source code is publicly available for broader research application.