EfficientNet-resDDSC: A Hybrid Deep Learning Model Integrating Residual Blocks and Dilated Convolutions for Inferring Gene Causality in Single-Cell Data

期刊论文
发表日期：

2024年11月22日

查看全文

Aimin Li,
Mingyue Li,
Rong Fei,
Saurav Mallik,
Bo Hu,
Yue Yu

摘要

Gene Regulatory Networks (GRNs) reveal complex interactions between genes in organisms, crucial for understanding the life system's operation. The rapid development of biotechnology, especially single-cell RNA sequencing (scRNA-seq), has generated a large amount of scRNA-seq data, which can be analyzed to explore the regulatory relationships between genes at the single-cell level. Previous models used to construct GRNs mainly aim at constructing associative relationships between genes, but usually fail to accurately reveal the causality between genes. Therefore, we present a hybrid deep learning model called EfficientNet-resDDSC (the EfficientNet with Residual Blocks and Depthwise Separable Dilated Convolutions) to infer causality between genes. The model inherits the basic structure of EfficientNet-B0 and incorporates residual blocks as well as dilated convolutions. The model's ability to extract low-level features at the primary stage is enhanced by introducing residual blocks. The model combines Depthwise Separable Convolution (DSC) in the inverted linear bottleneck layers with the dilated convolutions to expand the model's receptive fields without increasing the computational effort. This design enables the model to comprehensively reveal potential relationships among different genes in high-dimensional and high-noise single-cell data. In comparison with the five existing deep learning network models, EfficientNet-resDDSC’s overall performance is significantly better than others on four datasets. In this study, EfficientNet-resDDSC was further applied to construct GRNs for breast cancer patients, focusing on the related regulatory genes of the key gene BRCA1, which contributes to the advancement of breast cancer research and treatment strategies.

Graphical Abstract

The architectural framework of EfficientNet-resDDSC. The framework provides the basic processes of the EfficientNet-resDDSC model for inferring gene pair causality. These processes include the construction of a two-dimensional histogram of gene pairs $\left(a,b\right)$ $(a, b)$ , the underlying network architecture of the model, and the final output results. In the data input stage, a 32 × 32 two-dimensional histogram is constructed based on the expression values of each gene in the gene expression profile, which serves as the input data for the EfficientNet-resDDSC model. EfficientNet-resDDSC demonstrated the detailed network architecture of the model, which includes a convolutional layer, a maximal pooling layer, residual blocks, MBConv blocks, and for the MBDilatedConv blocks that incorporates dilated convolution. In this paper, we used the three-fold cross validation method, for the division of the dataset. The model was trained to obtain one probability vector [ ${p}_{0}$ $p_{0}$ , ${p}_{1}$ $p_{1}$ , ${p}_{2}$ $p_{2}$ ] for each pair of genes $\left(a,b\right)$ $(a, b)$ , where ${p}_{0}$ $p_{0}$ represents the probability that genes $a$ $a$ and $b$ $b$ have no regulatory relationship, ${p}_{1}$ $p_{1}$ represents the probability that gene $a$ $a$ regulates gene $b$ $b$ , and ${p}_{2}$ $p_{2}$ represents the probability that gene $b$ $b$ regulates gene $a$ $a$ . In causality prediction, for each pair of genes $\left(a,b\right)$ $(a, b)$ , we computed ${p}_{2}\left(a,b\right)$ $p_{2} (a, b)$ - ${p}_{2}\left(b,a\right)$ $p_{2} (b, a)$ as the pseudo-probability that $b$ $b$ regulates $a$ $a$ . The causal regulatory relationships of gene pairs were obtained by calculating the pseudo-probabilities of the gene pairs. This pseudo-probability was used to reflect the relative strength of regulation between two genes.