geneRFinder: gene finding in distinct metagenomic data complexities.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Metagenomics enables studying microbial communities and their functions through gene analysis.
• Next-generation sequencing generates large datasets, posing computational challenges for gene prediction.
• Current gene predictors struggle with metagenomic data complexity and lack standardized benchmarks.

Purpose of the Study:

• To develop an advanced gene prediction tool for complex metagenomic datasets.
• To establish a comprehensive benchmark dataset for evaluating gene prediction accuracy in metagenomics.

Main Methods:

• Developed geneRFinder, a machine learning-based gene predictor utilizing a pre-trained Random Forest model.
• Created a novel, comprehensive benchmark dataset for gene prediction based on the Critical Assessment of Metagenome Interpretation (CAMI) challenge.

Main Results:

• geneRFinder demonstrated superior performance compared to state-of-the-art tools like Prodigal and FragGeneScan.
• Achieved average prediction rate improvements of 54% over Prodigal and 64% over FragGeneScan.
• Showcased significantly higher specificity rates, outperforming FragGeneScan by 79 percentage points and Prodigal by 66 percentage points.

Conclusions:

• geneRFinder offers a robust solution for gene prediction across diverse metagenomic complexities.
• The provided benchmark dataset facilitates standardized evaluation of gene prediction tools.
• Both geneRFinder and the benchmark dataset are publicly available to advance metagenomic research.