| Description: |
Attempts to use machine learning (ML) for species identification of macrofungi have usually involved the use of image recognition to deduce the species from photographs, sometimes combining this with collection metadata. Our approach is different: we use a set of quantified morphological characters (for example, the average length of the spores) and locality (GPS coordinates). Using this data alone, the machine can learn to differentiate between species.Our case study is the genus Hebeloma, fungi within the order Agaricales, where species determination is renowned as a difficult problem. Whether it is as a result of recent speciation, the plasticity of the species, hybridization or stasis is a difficult question to answer. What is sure is that this has led to difficulties with species delimitation and consequently a controversial taxonomy.The Hebeloma Project—our attempt to solve this problem by rigorously understanding the genus—has been evolving for over 20 years. We began organizing collections in a database in 2003. The database now has over 10,000 collections, from around the world, with not only metadata but also morphological descriptions and photographs, both macroscopic and microscopic, as well as molecular data including at least an internal transcribed spacer (ITS) sequence (generally, but not universally, accepted as a DNA barcode marker for fungi (Schoch et al. 2012)), and in many cases sequences of several loci. Included within this set of collections are almost all type specimens worldwide. The collections on the database have been analysed and compared. The analysis uses both the morphological and molecular data as well as information about habitat and location. In this way, almost all collections are assigned to a species. This development has been enabled and assisted by citizen scientists from around the globe, collecting and recording information about their finds as well as preserving material.From this database, we have built a website, which updates as the database updates. The website ... |