Integrating Prior Knowledge Using Transformer for Gene Regulatory Network Inference.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory network (GRN) inference aims to uncover gene regulation rules from experimental data.
• Current methods face challenges in generalizing to diverse cell types and identifying novel regulators.

Purpose of the Study:

• To develop a novel framework, GRNPT, for robust and generalizable GRN inference.
• To overcome limitations of existing methods in handling diverse cell types and unseen regulators.

Main Methods:

• GRNPT utilizes a Transformer-based architecture integrating large language model (LLM) embeddings.
• It incorporates a temporal convolutional network (TCN) autoencoder to analyze single-cell RNA sequencing (scRNA-seq) trajectories.
• The framework leverages publicly available biological data for LLM embedding generation.

Main Results:

• GRNPT significantly outperforms existing supervised and unsupervised GRN inference methods.
• The framework demonstrates superior performance, especially with limited training data.
• GRNPT shows exceptional generalizability, accurately predicting regulatory relationships in novel cell types and for unseen regulators.

Conclusions:

• GRNPT offers a more accurate and comprehensive approach to understanding gene regulatory dynamics.
• The integration of LLM knowledge and deep learning enhances GRN inference capabilities.
• This framework advances the discovery of new regulatory rules and biological insights.