Fangorn Forest (F2)

Fangorn Forest (F2) classifies Geminiviridae genera and predicts genes ab initio from genomic sequences to enable taxonomic assignment and gene identification in geminivirus research.

Key Features:

• Ab initio gene prediction: Predicts genes using only genomic sequences and attributes derived from open reading frames (ORFs).
• Genus-level classification: Classifies genomes across nine Geminiviridae genera and their associated satellite DNAs using features extracted from complete genomes.
• Machine learning algorithms: Implements Support Vector Machines (SVM), Random Forest (RF), and Multilayer Perceptron models for predictive classification.
• Random Forest performance (genus): Achieved precision 0.966, recall 0.964, and AUC 0.995 for genus classification.
• Random Forest performance (gene): Achieved precision 0.983, recall 0.983, and AUC 0.998 for gene classification.
• Relevance to complex genomes: Applied to Geminiviridae genomes that include splicing mechanisms (e.g., Mastrevirus).
• Relevance to sequencing approaches: Relevant to datasets arising from rolling circle amplification (RCA) and metagenomics studies.

Scientific Applications:

• Geminiviridae taxonomy: Assigns genera within the family Geminiviridae for taxonomic and epidemiological studies.
• Gene identification: Predicts coding sequences and classifies ORFs in geminivirus genomes, including those with splicing complexity.
• Satellite DNA classification: Classifies associated satellite DNAs linked to Geminiviridae genomes.
• Plant virome and metagenomics analysis: Supports interpretation of plant virome surveys and metagenomic datasets involving geminiviruses.

Methodology:

Two training datasets were built (genus classification from complete genomes; gene classification from ORF-derived attributes) and three machine learning algorithms (SVM, Random Forest, Multilayer Perceptron) were applied, with Random Forest achieving the reported precision, recall, and AUC metrics.