Alfie

Alfie performs alignment-free taxonomic classification of cytochrome c oxidase subunit I (COI) DNA barcode sequences to distinguish target and off-target sequences for biodiversity and environmental DNA analyses.

Key Features:

• Alignment-Free Classification: Uses k-mer frequency profiles to classify DNA sequences without performing sequence alignment to reference databases.
• K-mer Feature Encoding: Computes k-mer frequencies (default k = 4) as input features for classification.
• Neural Network Classifier: Applies a neural network to k-mer frequency data to predict taxonomic kingdom assignments.
• Training Data: Classifier was trained and tested on approximately 58,000 publicly available COI sequences.
• High Accuracy and Generalization: Demonstrates greater than 99% accuracy for kingdom-level assignment and generalizes to sequences from previously unseen taxonomic classes.
• Target versus Off-target Discrimination: Identifies COI target sequences and separates off-target sequences without relying on existing reference barcodes, reducing bias against novel or rare genetic variants.
• API Extensibility: Provides a programmable interface to adapt the approach to other genes, barcodes, and taxonomic levels.

Scientific Applications:

• Biodiversity Estimation: Improves accuracy of diversity estimates from environmental DNA and bulk metabarcoding datasets by isolating target COI sequences.
• eDNA and Metabarcoding Noise Reduction: Removes off-target noise to increase reliability of downstream ecological and taxonomic analyses.
• Kingdom-Level Taxonomic Profiling: Enables rapid kingdom-level assignment to support ecological and evolutionary studies of species distribution and composition.

Methodology:

Alfie computes k-mer frequency vectors (default k = 4) from COI sequences and inputs them into a neural network classifier for alignment-free, kingdom-level taxonomic prediction; the model was trained and tested on ~58,000 publicly available COI sequences.