Species determination using AI machine-learning algorithms: Hebeloma as a case study.

Area of Science:

• Mycology
• Bioinformatics
• Computational Biology

Background:

• The fungal genus Hebeloma presents significant challenges in species identification.
• Traditional dichotomous keys have shown variable success rates for Hebeloma classification.
• A comprehensive database integrating morphological, micromorphological, and DNA sequence data is crucial for accurate identification.

Purpose of the Study:

• To develop an Artificial Intelligence (AI) machine-learning model for automated Hebeloma species identification.
• To leverage a large, curated database of Hebeloma collections for training and validation.
• To improve the accuracy and efficiency of species determination within the Hebeloma genus.

Main Methods:

• Compilation of a database containing approximately 9000 Hebeloma collections, including type specimens.
• Inclusion of detailed metadata, parametrized morphological descriptions, micromorphological analyses, and DNA sequences.
• Development of an AI machine-learning model utilizing locality data and morphological parameters for species identification.

Main Results:

• The AI species identifier achieved 77% correct identification based on the highest probability match.
• The model correctly identified 96% of species within its top three probabilistic determinations.
• Over 99% of collections were accurately classified within the top five most likely determinations using a validation set of over 600 collections.

Conclusions:

• The developed AI machine-learning tool demonstrates high accuracy in identifying Hebeloma species.
• This AI-driven approach offers a promising solution to the long-standing challenges in Hebeloma taxonomy.
• The integration of molecular and morphological data within a machine-learning framework significantly enhances species determination.