Gene knockout inference with variational graph autoencoder learning single-cell gene regulatory networks.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene function prediction is crucial for understanding biological systems.
• Experimental gene knockout (KO) studies are resource-intensive and ethically challenging.
• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data.

Purpose of the Study:

• To introduce Gene Knockout Inference (GenKI), a novel computational tool for predicting gene function.
• To enable gene function prediction using only wild-type (WT) scRNA-seq data, bypassing the need for KO samples.
• To provide a robust and scalable framework for in-silico gene function studies.

Main Methods:

• GenKI utilizes a variational graph autoencoder (VGAE) model.
• It learns gene representations and interactions from WT scRNA-seq data and a single-cell gene regulatory network (scGRN).
• Virtual KO data is generated by computationally removing gene edges from the scGRN, and differences are analyzed in the latent space.

Main Results:

• GenKI accurately approximates gene KO perturbation profiles in simulations.
• The tool outperforms state-of-the-art methods under various evaluation conditions.
• GenKI successfully recapitulates findings from real KO experiments and predicts cell type-specific gene functions.

Conclusions:

• GenKI offers a powerful in-silico alternative to experimental KO studies.
• The method can reduce reliance on genetically modified animals or perturbed systems.
• GenKI facilitates efficient and scalable gene function discovery using scRNA-seq data.