Software for predicting microbial CO₂ fixation pathways developed

A joint research group consisting of Associate Professor Hiroshi Mori from the National Institute of Genetics, Research Organization of Information and Systems (ROIS) and researchers from the Database Center for Life Science (DBCLS) at the ROIS Joint Support-Center for Data Science Research, the National Institute of Technology and Evaluation (NITE), and OKBP Inc. has developed AutoFixMark, a software tool that predicts the CO₂ fixation pathways present in chemolithoautotrophic bacteria from genomic data with high accuracy. The tool enables efficient screening of microorganisms capable of utilizing CO₂ as a resource from vast microbial genome datasets. It is expected to advance our understanding of the global carbon cycle and contribute to sustainable biomanufacturing through biotechnology. The results were published in Scientific Data.

Microbial CO₂ fixation is an essential process that allows microorganisms to survive in carbon-limited environments, and it plays a key role in the global carbon cycle. The CO₂ fixation pathways found in chemolithoautotrophic bacteria are diverse: seven distinct types are known, including the Calvin-Benson-Bassham (CBB) cycle.

However, because the enzyme genes involved in these pathways are present across diverse lineages and some enzymes participate in multiple pathways, accurately identifying which pathways a given organism possesses from genomic data alone has been difficult. Existing metabolic pathway prediction tools for bacteria are useful for predicting common metabolic pathways. However, they have faced issues with accuracy when predicting the diverse CO₂ fixation pathways—particularly some that were discovered relatively recently.

The research group therefore used genome data and literature information from 15 representative chemolithoautotrophic bacteria to define, for all seven known CO₂ fixation pathways, the marker enzymes essential for identifying each pathway and the corresponding KEGG Orthology (KO) IDs for their genes. To accommodate the diversity of enzymes, flexible logical rules were established, such as "at least one gene from this set is sufficient" or "all genes forming a complex are required."

Based on the marker enzymes and rules they had defined, the group then developed AutoFixMark, a tool that automatically determines whether each CO₂ fixation pathway is present or absent from a genome's gene list (KO composition).

To properly evaluate the tool's performance, the NITE group took the lead in manually curating the genomic information of 347 microbial strains (representing 16 phyla) whose CO₂ fixation capabilities had been confirmed through literature surveys, thereby building a reference dataset. They used this dataset as ground truth to compare the prediction accuracy of AutoFixMark against existing tools (METABOLIC and gapseq). AutoFixMark demonstrated high predictive accuracy even for pathways that had been difficult to predict with existing tools, such as the dicarboxylate/4-hydroxybutyrate (DC/4HB) cycle and the reductive glycine (rGly) pathway.

Since AutoFixMark can predict the presence or absence of CO₂ fixation pathways from genomic sequence data alone, it is easily applicable to metagenomic datasets and single-cell genome data that include microorganisms that are difficult to culture. Beyond helping to elucidate the phylogenetic diversity of autotrophic bacteria in the environment, it is expected to serve as a foundation for biotechnology applications aimed at realizing a decarbonized society, including aiding the search for microorganisms that can produce useful substances using CO₂ as a feedstock.

All curated enzyme gene sets, prediction rules, the AutoFixMark software, and the benchmark datasets have been made publicly available through the GitHub repository and the data repository Zenodo, making them accessible resources for researchers worldwide.

Journal Information
Publication: Scientific Data
Title: A curated resource of chemolithoautotrophic genomes and marker genes for CO₂ fixation pathway prediction
DOI: 10.1038/s41597-026-06655-z