Exploring soybean metabolic pathways based on probabilistic graphical model and knowledge-based methods

Jie Hou, Gary Stacey, Jianlin ChengEURASIP J. Bioinform. Syst. Biol 2015 · OpenReview

TL;DR

This paper presents a novel method for reconstructing soybean metabolic pathways by integrating gene expression data with existing biological knowledge using Bayesian networks. This approach significantly improves upon traditional methods by predicting more accurate and comprehensive metabolic pathways, uncovering previously unknown gene relationships.

What Problem Does This Paper Address?

Understanding the intricate metabolic pathways within an organism is crucial for deciphering its biological functions and potential applications. Soybean, a vital crop for food and industry, has had its genome sequenced, leading to abundant RNA-seq data. However, existing methods for reconstructing metabolic pathways often rely on simple sequence homology mapping, which can miss complex interactions and relationships. This paper aims to address this gap by developing a more sophisticated approach that leverages both data-driven (gene expression) and knowledge-driven (existing pathway databases and interaction predictions) methods to build more accurate and complete metabolic pathway maps for soybean.

Methods

The researchers developed a multi-step workflow for metabolic pathway reconstruction. Key steps include:

  • Data Preprocessing: This involved removing non-expressed genes from RNA-seq data and generating protein sequences from annotated soybean genomes.
  • Initial Pathway Construction: Protein sequences were mapped to known metabolic pathways in the KEGG database using the KAAS server. This provided an initial, albeit potentially incomplete, set of pathway relationships.
  • Gene Clustering: Genes with similar expression patterns across different tissues were grouped using an Expectation-Maximization (EM) clustering algorithm. This step assumes that genes with similar expression profiles are likely involved in the same metabolic processes.
  • Relation Knowledgebase Construction: Information on existing relationships between KEGG Orthology (KO) numbers was compiled. This knowledgebase was used to create a "whitelist" of known relationships and a "blacklist" of unsupported relationships to guide the Bayesian network construction.
  • Bayesian Network Pathway Construction: A Bayesian network was employed to model probabilistic relationships between genes. This method, combined with the knowledgebase constraints (whitelist and blacklist), allowed for the prediction of new gene-gene interactions and the expansion of initial pathways. The process iteratively built the network, incorporating gene clusters and predicted relationships.
  • Cycle Detection: A crucial step was implemented to detect and prevent cycles in the Bayesian network, as these are not directly handled by the model.
  • Functional Enrichment Analysis: To validate the predicted pathways, Gene Ontology (GO) terms were used to assess the functional coherence of predicted gene relationships.

Datasets

  • Soybean RNA-seq data: This dataset comprised gene expression measurements from 14 different soybean tissue-specific conditions.
  • Soybean Genome Sequence: Used as the basis for gene annotation and protein sequence extraction.
  • KEGG Database: Provided reference metabolic pathways and KO numbers for initial mapping and validation.
  • Soybean Knowledge Base (SoyKB): Used to extract protein sequences for annotated soybean genes.

Key Insights and Ideas

  • Synergy of Data and Knowledge: The core idea is that combining gene expression data with existing biological knowledge (KEGG pathways, predicted interactions) leads to more robust and accurate pathway reconstruction than either approach alone.
  • Bayesian Networks for Discovery: Bayesian networks are powerful tools for inferring probabilistic relationships and discovering novel interactions that might be missed by simpler homology-based methods.
  • Gene Expression as a Proxy for Function: The assumption that genes with similar expression patterns are functionally related is a key driver for using gene clustering to expand pathway predictions.
  • Importance of Knowledge Constraints: Using a whitelist and blacklist derived from existing knowledge significantly prunes the search space for Bayesian network learning, improving efficiency and accuracy.
  • Predicting the Unknown: The method not only confirms known relationships but also successfully predicts new gene relationships not present in the initial KEGG mapping.

Key Results

  • The proposed method demonstrated improved recall rates compared to traditional KEGG mapping, indicating that it correctly identifies a larger proportion of true edges in target pathways.
  • The method also showed higher precision rates, meaning that the predicted edges are more likely to be correct.
  • The functional enrichment analysis supported the validity of the predicted pathways, showing that genes linked by the Bayesian network often share similar GO terms.
  • The study successfully reconstructed metabolic pathways, including the glyoxylate and dicarboxylate metabolism pathway (KO00630), highlighting the method's ability to capture complex interactions.

Limitations and Future Work

  • Computational Cost: While knowledge constraints were used to improve efficiency, Bayesian network learning can still be computationally intensive.
  • Data Quality: The accuracy of the reconstruction is dependent on the quality and completeness of the input RNA-seq data and the existing KEGG knowledgebase.
  • Future Work: The authors suggest that this data- and knowledge-driven probabilistic method can be generally applicable to other genomes where similar starting data are available, implying potential for broader application in systems biology. Further validation and refinement of the method for different organisms and pathway types would be beneficial.

Summary updated

Abstract

Soybean (Glycine max) is a major source of vegetable oil and protein for both animal and human consumption. The completion of soybean genome sequence led to a number of transcriptomic studies (RNA-seq), which provide a resource for gene discovery and functional analysis. Several data-driven (e.g., based on gene expression data) and knowledge-based (e.g., predictions of molecular interactions) methods have been proposed and implemented. In order to better understand gene relationships and protein interactions, we applied probabilistic graphical methods, based on Bayesian network and knowledgebase constraints using gene expression data to reconstruct soybean metabolic pathways. The results show that this method can predict new relationships between genes, improving on traditional reference pathway maps.
BibTeX
@article{hou2015exploring,
  author = {Jie Hou and Gary Stacey and Jianlin Cheng},
  title = {{Exploring soybean metabolic pathways based on probabilistic graphical model and knowledge-based methods}},
  journal = {EURASIP J. Bioinform. Syst. Biol.},
  year = {2015},
  url = {https://openreview.net/forum?id=wVFXmeCIQE},
}
Figure