DeepCE: a deep learning framework for correlation-enhanced gene regulatory network inference in single-cell RNA...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals gene expression dynamics and cellular heterogeneity.
• Deep learning (DL) shows promise for inferring genetic regulation but struggles with complex mechanisms.
• New algorithms are needed to improve the effectiveness and reliability of gene regulatory network (GRN) inference.

Purpose of the Study:

• To introduce DeepCE, a novel DL framework designed for correlation-enhanced GRN inference.
• To improve the accuracy and robustness of GRN inference by integrating advanced DL techniques.

Main Methods:

• DeepCE integrates bidirectional gated recurrent units (GRUs) with convolutional neural networks (CNNs).
• Bidirectional GRUs capture dynamic temporal dependencies in gene expression data.
• CNNs analyze local spatial patterns within scRNA-seq data to uncover complex gene-gene interactions.

Main Results:

• DeepCE enhances the extraction of dynamic gene regulation.
• The framework smooths noisy gene expression data, extracts time-lagged regulatory signals, and filters spurious correlations.
• Experiments on mouse and human datasets show DeepCE outperforms existing methods, achieving superior AUROC and AUPR scores.

Conclusions:

• DeepCE provides a powerful and reliable framework for high-quality GRN inference.
• The proposed method advances the understanding of gene regulatory mechanisms from single-cell data.
• DeepCE offers improved accuracy and robustness compared to current state-of-the-art approaches.