InfoSEM: A Deep Generative Model with Informative Priors for Gene Regulatory Network Inference

2025-03-06

AI Summary

Supervised gene regulatory network (GRN) inference methods rely on costly, imbalanced, and gene-biased ground-truth (GT) labels that poorly reflect true biological regulation. To address this, we propose the first unsupervised generative GRN inference framework: it incorporates biologically grounded text embeddings (e.g., PubMedBERT) as semantic priors and integrates heterogeneous biological knowledge; jointly models gene expression and regulatory topology via a variational autoencoder coupled with differentiable graph learning; and introduces a biology-driven evaluation system aligned with downstream tasks (e.g., biomarker discovery) to expose implicit biases in supervised approaches. On four benchmark datasets, our method achieves an average 38.5% performance gain over state-of-the-art supervised baselines; incorporating weak label priors further improves performance by 11.1%; and critically mitigates both label bias and class imbalance—key limitations of GT-dependent methods.

Technology Category

• Reasoning under Uncertainty: Graphical Models
• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Inferring Gene Regulatory Networks (GRNs) from gene expression data is crucial for understanding biological processes. While supervised models are reported to achieve high performance for this task, they rely on costly ground truth (GT) labels and risk learning gene-specific biases, such as class imbalances of GT interactions, rather than true regulatory mechanisms. To address these issues, we introduce InfoSEM, an unsupervised generative model that leverages textual gene embeddings as informative priors, improving GRN inference without GT labels. InfoSEM can also integrate GT labels as an additional prior when available, avoiding biases and further enhancing performance. Additionally, we propose a biologically motivated benchmarking framework that better reflects real-world applications such as biomarker discovery and reveals learned biases of existing supervised methods. InfoSEM outperforms existing models by 38.5% across four datasets using textual embeddings prior and further boosts performance by 11.1% when integrating labeled data as priors.

Problem

• Inferring Gene Regulatory Networks without costly ground truth labels.
• Leveraging textual gene embeddings to improve GRN inference accuracy.
• Proposing a benchmarking framework to evaluate real-world GRN applications.

Innovation

• Unsupervised generative model with textual gene embeddings
• Integrates ground truth labels to avoid biases
• Biologically motivated benchmarking framework for real-world applications