HFTC: a hierarchical fungal taxonomic classification model for ITS sequences using low-dimensional embedding features

Area of Science:

• Mycology
• Bioinformatics
• Computational Biology

Background:

• Fungal identification via ITS sequencing is crucial for ecological and biodiversity research.
• Existing methods struggle with high-dimensional data and taxonomic inconsistencies.

Purpose of the Study:

• To develop a novel hierarchical fungal taxonomic classifier named HFTC.
• To improve the accuracy and computational efficiency of fungal classification.

Main Methods:

• Implemented a multi-level random forest (RF) architecture for HFTC.
• Utilized a bidirectional k-mer strategy and Word2Vec embedding to reduce feature dimensionality from 4^k to 200.
• Captured contextual sequence information from both orientations.

Main Results:

• HFTC achieved a Matthews correlation coefficient (MCC) of 95.31% and overall accuracy (ACC) of 95.25%.
• Species-level hierarchical accuracy (HA) reached 95.10%, outperforming the deep learning baseline CNN-Duong by 3.2%.
• HFTC demonstrated superior hierarchical consistency with the smallest ACC-HA discrepancy (1.60%).

Conclusions:

• HFTC provides a scalable and accurate solution for fungal taxonomic classification.
• Its compact feature representation and hierarchical design are ideal for microbial diversity research.
• The source code and datasets are publicly available.