funbarRF: DNA barcode-based fungal species prediction using multiclass Random Forest supervised learning model.

Area of Science:

• Mycology
• Bioinformatics
• Conservation Biology

Background:

• Accurate identification of fungal species is crucial for conserving fungal diversity.
• Morphological identification is challenging for uncultured fungi.
• DNA barcoding, specifically the internal transcribed spacer (ITS) region of ribosomal DNA (rDNA), offers a viable alternative for species identification.

Purpose of the Study:

• To develop a computational method for identifying unknown fungal species using DNA barcode sequences.
• To address the challenges posed by high variability within ITS regions for accurate fungal species prediction.

Main Methods:

• A Random Forest (RF)-based predictor was developed for fungal species identification.
• Reference and query sequences were converted to numeric features based on gapped base pair compositions.
• The model was trained and tested using these numeric features.

Main Results:

• The RF model achieved over 85% accuracy with 4 sequences per species and stabilized at approximately 88% with ≥7 sequences per species in the reference set.
• The proposed model demonstrated comparable accuracy to existing methods via cross-validation.
• The model outperformed several existing models in identifying non-fungal species.

Conclusions:

• An online prediction server, "funbarRF", and an R-package "funbarRF" have been developed for fungal species identification using DNA barcodes.
• These tools facilitate high-throughput sequence data analysis for fungal taxonomy.
• This work supports future efforts in fungal taxonomy assignments based on DNA barcoding.