Ziwei Yang, Ph.D.
01 · Model Biology · Mechanistic Biological Modeling

DBgDel

Using known gene deletion strategies to speed up the search for new ones

DBgDel takes the gene deletion strategies already stored in the MetNetComp database and uses them to narrow the search space of a gene deletion algorithm, so that strategies for new target metabolites are found faster.

Metabolic ModelingMetabolic Engineering Ziwei Yang, Takeyuki Tamura · IEEE Transactions on Computational Biology and Bioinformatics, 2025
DBgDel framework. STEP 1: known gene deletion strategies from the MetNetComp database give the shared remaining genes of the model, which define a sub-network that enables growth-coupled production. STEP 2: these genes initialise the gDel_minRN algorithm, which searches a narrowed space and returns a metabolic network that couples growth with target production.
DBgDel at a glance. STEP 1 extracts the genes that known maximal deletion strategies leave in place; STEP 2 uses them as the starting gene pool of the MILP-based gDel_minRN algorithm, which then searches a much smaller space for a new target metabolite. Figure adapted from Yang & Tamura, IEEE TCBB 2025.

Overview

Designing a strain for growth-coupled production means finding a set of genes to delete in a genome-scale metabolic model so that the cell can only grow while making the target compound. The number of possible deletions grows quickly with model size: on the genome-scale iML1515 model, the previous best method, gDel_minRN, needed about 2,500 seconds per target metabolite on average.

Databases such as MetNetComp already store computed strategies for many metabolites. DBgDel treats them as prior knowledge. Genes that none of the stored maximal strategies delete mark a core sub-network that supports growth-coupled production, so they can be kept from the start.

The search then only has to decide the remaining genes. The paper also reports a small e_coli_core test that plugs a different algorithm, gMCSE, into the same framework.

Method

Two steps, shown in the figure above.

  1. 01

    Extract the shared remaining genes

    Collect the maximal deletion strategies stored for the model's other metabolites and keep the genes that none of them delete. These form the initial remaining gene pool, which defines a sub-network that can support growth-coupled production.

  2. 02

    Search in the narrowed space

    Run an extended gDel_minRN that starts from this gene pool, so the MILP only decides the other genes. Candidates are checked by maximizing growth (GR) and evaluating target production (PR).

Key Results

Against gDel_minRN, the previous best method

5.89×
faster on iML1515
431.75 s vs 2,541.23 s per target
507/991
iML1515 targets solved
gDel_minRN: 508/991
1.45×
faster on e_coli_core
1.12 s vs 1.62 s per target

On iML1515, GDLS, gMCSE and optGene solved none of the 991 targets. A target counts as solved when growth rate and production rate both reach at least 0.001 under growth maximization.

Benchmark

Success rate and run time against four methods DBgDelOther methods
e_coli_core 40 targets
DBgDel 24/40 1.12 s
GDLS 5/40 1.24 s
gMCSE 7/40 1.73 s
optGene 22/40 1779.30 s
gDel_minRN 28/40 1.62 s
iMM904 700 targets
DBgDel 81/700 79.92 s
GDLS 0/700 243.42 s
gMCSE 17/700 584.24 s
optGene 27/700 1986.45 s
gDel_minRN 116/700 422.32 s
iML1515 991 targets
DBgDel 507/991 431.75 s
GDLS 0/991 244.58 s
gMCSE 0/991 1892.39 s
optGene 0/991 2193.45 s
gDel_minRN 508/991 2541.23 s
A strategy counts as successful if growth rate and production rate both reach ≥ 0.001 during growth maximization; time is the average elapsed seconds per target (log scale). Non-extracellular target metabolites. Data from Tables VIII, X and XII of the paper (arXiv v3); redrawn.

Resources

Paper

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

IEEE TCBB 2025 · preprint on arXiv

Code

MetNetComp/DBgDel

Source code