geneRFinder

geneRFinder predicts protein-coding genes in metagenomic sequences using a Random Forest machine learning model to improve gene annotation accuracy in metagenomic datasets.

Key Features:

• Machine Learning Approach: Uses a pre-trained Random Forest ensemble model to classify coding versus non-coding regions in metagenomic data.
• Performance Comparisons: Achieves average prediction rates 54% higher than Prodigal and 64% higher than FragGeneScan, with specificity improvements of 66 percentage points over Prodigal and 79 percentage points over FragGeneScan.
• Statistical Validation: Performance differences were validated using McNemar’s test with a 99% confidence level across datasets.
• Benchmark Dataset: Trained and evaluated on a CAMI-derived benchmark dataset containing labeled gene-region data for metagenomic evaluation.
• Next-Generation Sequencing Handling: Designed to address the computational demands posed by next-generation sequencing technologies in metagenomic analyses.

Scientific Applications:

• Metagenomics Research: Improves gene prediction and annotation in complex metagenomic datasets.
• Microbial Ecology: Enables analysis of microbial community composition and functional potential via more accurate gene calls.
• Environmental Microbiology and Ecology: Supports studies of environmental samples by providing higher-confidence gene-level information.
• Method and Tool Benchmarking: Provides a reference approach and CAMI-derived labeled data for benchmarking gene prediction tools.

Methodology:

A pre-trained Random Forest model was trained on a CAMI-derived benchmark of labeled gene regions to classify coding versus non-coding regions and annotate genes in metagenomic sequences, with performance assessed using McNemar’s test.