Construction of reference library for MALDI-based identification that enable species-level to drug-resistant strain discrimination.

Author:

Sayaka Ban (JP)

Abstract:

Objectives: MALDI offers a cost-effective and time-saving solution for clinical laboratory identification, alternative to sequencing. Fungal identification, however, faces challenges due to limited references compared to bacteria. For address this, we have previously constructed a library using supervised machine learning for Aspergillus section Nigri, a discriminative model for identifying drug-resistant strains of A. fumigatus, and a main spectral profiles library containing a wide range of species. Here, we presented the evaluation results of the three resources.
Methods: A collaborative project among the three microbe culture collections has resulted in the development EMALiMB: an extended database for rapid identification by the Biotyper (Bruker). The library includes a total of 633 species from 153 genera of fungi, 1,709 MSPs. For section Nigri, SARAMIS (bioMérieux) was used for discrimination among species. We also created a model for identification azole resistant A. fumigatus strains by discriminative methods and multivariate analysis (eMSTAT, Shimadzu). Validation test was conducted by using novel isolates, respectively.
Results: The European reference had previously resulted in lower scores for isolates from Japan, but they improved by add-on EMALiMB, especially for non-albicans species. This unsupervised machine learning library effectively distinguished closely related species within Aspergillus section Fumigati, but not within section Nigri. With the supervised machine learning library, it was possible to differentiate closely related species such as between A. tubingensis and A. neoniger. However, it struggled with A. welwitschiae from A. niger s.s. as shown by genomic analyses in the previous study (Bian et al. 2022, Stud Mycol). This system can accommodate a wide range of fungal species but may require redesigning from scratch due to enhancing the evaluation of species-specific peaks, when changes occur in the classification system or new species emerge. The discriminative model for azole-resistant A. fumigatus was achieving a 97.06% accuracy rate in VITEK MS and 92.5% in Biotyper for clinical isolates. However, false positives were frequent in isolates from tulip bulb (TR variant), suggested that this discriminative model might reflect regional characteristics rather than resistant characteristics.
Conclusions: These MALDI resources will significantly improve rapid clinical identification, though they may need adjustments to libraries and models to align with specific applications.

Abstract Number: 26

Conference Year: 2024

Conference abstracts, posters & presentations

Showing 10 posts of 17325 posts found.

  • Title

    Author

    Year

    Number

    Poster