DBgDel: Database-Enhanced Gene Deletion Framework for Growth-Coupled Production in Genome-Scale Metabolic Models

Ziwei Yang, Takeyuki TamuraIEEE Trans. Comput. Biol. Bioinform 2025 · DOI

TL;DR

This paper introduces DBgDel, a novel computational framework that significantly speeds up the process of identifying gene deletions for optimizing microbial production. By leveraging existing databases of successful gene deletion strategies, DBgDel intelligently narrows down the search space, leading to a 6.1-fold average acceleration in computation time while maintaining a high success rate.

What Problem Does This Paper Address?

Metabolic engineering aims to engineer microorganisms to produce valuable compounds. A key strategy is to modify a microbe's metabolic network by deleting specific genes. The goal is often "growth-coupled production," where cell growth and the production of a target metabolite occur simultaneously. Simulating these gene deletions in large, genome-scale metabolic models is computationally intensive, especially when searching for optimal deletion strategies for numerous potential target metabolites. Existing methods often recalculate strategies from scratch for each new target, wasting computational resources by not leveraging prior knowledge. This paper addresses this inefficiency by proposing a database-enhanced approach.

Methods

DBgDel is a two-step framework:

  • STEP 1: Extracting Prior Information (Database Mining): DBgDel first queries a database (MetNetComp) containing known gene deletion strategies for growth-coupled production. It identifies "maximal gene deletion strategies" (those that delete the most genes while still achieving the desired outcome) for existing metabolites. From these, it constructs a "remaining gene set" (Gremain), which represents genes that are consistently not deleted in successful strategies. This set acts as a prior knowledge base.
  • STEP 2: Database-Enhanced Gene Deletion Calculation: DBgDel then uses an extended version of the gDel_minRN algorithm. This algorithm is a mixed-integer linear programming (MILP) approach that aims to maximize repressed reactions and cell growth. The key innovation here is that DBgDel initializes gDel_minRN with the Gremain set from STEP 1. This "initial remaining gene pool" prevents the algorithm from considering deletions of genes within Gremain, effectively narrowing the search space and accelerating computation.

Datasets

The paper utilizes the MetNetComp database as a source of known gene deletion strategies. For evaluation, they use three established constraint-based metabolic models:

  • e_coli_core: A simplified model of E. coli metabolism.
  • iMM904: A genome-scale model of S. cerevisiae.
  • iML1515: A genome-scale model of E. coli.

Key Insights and Ideas

  • Leveraging Existing Knowledge is Key: The most significant insight is that pre-existing gene deletion strategies contain valuable information that can dramatically reduce computational burden. Instead of recalculating everything, we can use past successes to guide future searches.
  • "Remaining Genes" as a Powerful Prior: The concept of identifying genes that are consistently not deleted in successful strategies (Gremain) is a clever way to encode prior knowledge. This set acts as a constraint, preventing the algorithm from exploring unproductive search paths.
  • Trade-off Management: DBgDel demonstrates that it's possible to achieve a good balance between computational efficiency and success rate. While some methods might achieve slightly higher success rates, they come at a much higher computational cost. DBgDel offers a compelling compromise.
  • Scalability Matters: The performance advantage of DBgDel becomes more pronounced as the size and complexity of the metabolic models increase, highlighting its utility for large-scale engineering projects.

Key Results

  • Significant Speedup: DBgDel achieved an average of 6.1-fold acceleration in computational speed compared to existing methods across the tested models. For the largest model (iML1515), the speedup was even more substantial.
  • Competitive Success Rates: While not always achieving the absolute highest success rates, DBgDel's success rates were competitive, often closely matching or exceeding those of other methods, especially when compared to speed-optimized approaches. For example, in e_coli_core, DBgDel achieved a 24/40 success rate, outperforming GDLS (5/40) and gMCSE (7/40), and nearing gDel_minRN (28/40).
  • Model Size Impact: The performance gains were more significant for larger models like iMM904 and iML1515, where DBgDel's efficiency advantage over other methods was more pronounced.

Limitations and Future Work

  • Database Dependency: The effectiveness of DBgDel is dependent on the availability and quality of gene deletion strategies in the MetNetComp database. For models or metabolites with limited prior data, the benefits might be reduced.
  • "Predicted-Gremain" vs. "Gremain": The paper notes that using "Gremain genes" (derived from all metabolites) as the initial pool generally leads to higher success rates than "Predicted-Gremain genes" (derived from extracellular metabolites), but at a computational cost. The default "Predicted-Gremain" offers a good balance.
  • Future Directions: The authors suggest exploring how to reduce the search space for other gene deletion algorithms using their framework, demonstrating its broader applicability. They also aim to develop more flexible methods for extracting information from gene deletion databases, moving beyond handcrafted rules.

Summary updated

BibTeX
@article{yang2025dbgdel,
  author = {Ziwei Yang and Takeyuki Tamura},
  title = {{DBgDel: Database-Enhanced Gene Deletion Framework for Growth-Coupled Production in Genome-Scale Metabolic Models}},
  journal = {IEEE Trans. Comput. Biol. Bioinform.},
  year = {2025},
  doi = {10.1109/tcbbio.2025.3555999},
}
Figure