Weighted Fused Pathway Graphical Lasso for Joint Estimation of Multiple Gene Networks

Nuosi Wu, Jiang Huang, Xiao-Fei Zhang, Le Ou-Yang, Shan He, Zexuan Zhu, Weixin XieFrontiers in Genetics 2019 · DOI

TL;DR

This paper introduces a novel method, Weighted Fused Pathway Graphical Lasso (WFPGL), to infer gene regulatory networks (GRNs) by jointly analyzing multiple datasets (e.g., from different disease states or subtypes). WFPGL uniquely incorporates prior biological knowledge from gene interaction databases and known biological pathways to improve accuracy and interpretability.

What Problem Does This Paper Address?

Understanding how genes interact within a cell is crucial for deciphering biological processes and disease mechanisms. Gene Regulatory Networks (GRNs) are a powerful tool for this, often modeled using Gaussian Graphical Models (GGMs). However, traditional GGMs typically analyze one dataset at a time, ignoring valuable information that could be gleaned from comparing multiple datasets (e.g., healthy vs. diseased tissues, different cancer subtypes). Furthermore, these models often overlook existing biological knowledge, such as which genes are known to interact or which genes participate in specific biological pathways. This paper addresses the challenge of jointly inferring multiple GRNs while effectively leveraging both shared information across datasets and external biological prior knowledge.

Methods

The core of this work is the Weighted Fused Pathway Graphical Lasso (WFPGL) algorithm. It builds upon existing graphical lasso techniques but introduces several key innovations:

  • Joint Estimation of Multiple Networks: Instead of analyzing each dataset (or "state") independently, WFPGL estimates networks for multiple states simultaneously. This allows the model to learn commonalities and differences across states, leading to more robust inferences.
  • Pathway-Based Constraints: WFPGL incorporates prior knowledge about biological pathways. It assumes that genes not participating in the same pathway are less likely to interact. This constraint helps to prune spurious connections and focus on biologically plausible interactions.
  • Weighted Fused Lasso Penalty: This penalty term is designed to leverage prior gene interaction data. It encourages the model to identify differential edges (changes between states) that are more likely to occur between genes already known to interact. The "weighted" aspect allows for fine-tuning the influence of this prior information.
  • Optimization: The complex optimization problem is solved using the Alternating Direction Method of Multipliers (ADMM), a standard technique for distributed optimization. The authors also employ a "pathway separable" approach to decompose the problem into smaller, more manageable subproblems, accelerating computation.

Datasets

The paper uses both synthetic and real-world datasets for evaluation:

  • Synthetic Datasets: Generated to simulate gene expression data from two different states. These datasets were used to compare WFPGL against five other state-of-the-art methods under varying levels of prior information.
  • Real Datasets:
    • Type 2 Diabetes: Gene expression data from insulin-resistant (IR) and insulin-sensitive (IS) patients. Prior information included known gene interactions from the TRRUST database and pathways from KEGG.
  • Breast Cancer Subtypes: Gene expression data from four different breast cancer subtypes (basal-like, HER2-enriched, luminal A, and luminal B). Similar prior information sources were used.

Key Insights and Ideas

  • Prior Knowledge is Powerful: The study strongly demonstrates that incorporating biological prior knowledge (pathways and gene interactions) significantly improves the accuracy of GRN inference, especially for identifying differential networks.
  • Joint Analysis Outperforms Independent Analysis: By analyzing multiple states together, WFPGL can leverage shared information, leading to more accurate and robust network estimations compared to methods that analyze each state in isolation.
  • Pathway Constraints Guide Inference: The assumption that genes within the same pathway are more likely to interact helps to filter out noise and focus on biologically relevant connections, making the inferred networks more interpretable.
  • Differential Networks Reveal Disease Mechanisms: Identifying how gene networks change between different disease states or subtypes is key to understanding disease progression and finding potential therapeutic targets. WFPGL excels at this by highlighting these dynamic changes.

Key Results

  • Synthetic Data: WFPGL consistently outperformed five other state-of-the-art graphical models in both individual network estimation (measured by TPR/FPR) and differential network estimation (measured by TPDR/FPDR) across various scenarios, especially when prior information was available. The performance improvement was particularly noticeable with increasing levels of prior gene interaction information.
  • Real Data (Type 2 Diabetes): WFPGL identified several "hub genes" in the differential network between IR and IS patients that have known associations with insulin resistance, providing biologically relevant insights.
  • Real Data (Breast Cancer): The analysis of breast cancer subtypes revealed both shared and subtype-specific network structures, offering potential clues into the mechanisms of cancer differentiation and identifying candidate genes for further study.

Limitations and Future Work

  • Limitations:
    • The performance of WFPGL is dependent on the quality and completeness of the prior knowledge (pathways and gene interactions). Incomplete or inaccurate prior information could lead to suboptimal results.
  • The paper acknowledges that for real data, ground truth is unavailable, making definitive validation challenging. The assessment relies on identifying known biologically relevant genes.
  • Future Work:
    • The authors suggest that their "pathway" concept is flexible and could be extended. For instance, incorporating information about transcription factors (TFs) and their regulated genes could lead to more refined pathway definitions.
  • If pathway information is scarce or unknown, the model can degenerate to a weighted fused graphical lasso, suggesting its adaptability to different levels of prior knowledge.
  • The authors propose that combining TF information with gene interaction and pathway data could further enhance the model's capabilities.

Summary updated

Abstract

Gene regulatory networks (GRNs) are often inferred based on Gaussian graphical models that could identify the conditional dependence among genes by estimating the corresponding precision matrix. Classical Gaussian graphical models are usually designed for single network estimation and ignore existing knowledge such as pathway information. Therefore, they can neither make use of the common information shared by multiple networks, nor can they utilize useful prior information to guide the estimation. In this paper, we propose a new weighted fused pathway graphical lasso (WFPGL) to jointly estimate multiple networks by incorporating prior knowledge derived from known pathways and gene interactions. Based on the assumption that two genes are less likely to be connected if they do not participate together in any pathways, a pathway-based constraint is considered in our model. Moreover, we introduce a weighted fused lasso penalty in our model to take into account prior gene interaction data and common information shared by multiple networks. Our model is optimized based on the alternating direction method of multipliers (ADMM). Experiments on synthetic data demonstrate that our method outperforms other five state-of-the-art graphical models. We then apply our model to two real datasets. Hub genes in our identified state-specific networks show some shared and specific patterns, which indicates the efficiency of our model in revealing the underlying mechanisms of complex diseases.
BibTeX
@inproceedings{wu2019weighted,
  author = {Nuosi Wu and Jiang Huang and Xiao-Fei Zhang and Le Ou-Yang and Shan He and Zexuan Zhu and Weixin Xie},
  title = {{Weighted Fused Pathway Graphical Lasso for Joint Estimation of Multiple Gene Networks}},
  booktitle = {Frontiers in Genetics},
  year = {2019},
  doi = {10.3389/fgene.2019.00623},
}
Figure