News Release

Refining biome labeling for microbial community samples: AI approach unravels hidden

Peer-Reviewed Publication

Chinese Society for Environmental Sciences

Graphical abstract.

image: Graphical abstract. view more 

Credit: Environmental Science and Ecotechnology

In a groundbreaking study published on July 26, 2023, in the journal Environmental Science and Ecotechnology, researchers from Huazhong University of Science and Technology have introduced "Meta-Sorter," an AI-based method that leverages neural networks and transfer learning to significantly improve biome labeling for thousands of microbiome samples in the MGnify database, especially those with incomplete information. The Meta-Sorter approach comprises two crucial steps. Firstly, a neural network model is meticulously constructed using 118,592 microbial samples from 134 biomes and their respective biome ontology, boasting an impressive average AUROC of 0.896. This model accurately classifies samples with detailed biome information, serving as a strong foundation for further analyses. Secondly, to address the challenge of newly introduced samples with different characteristics, researchers incorporated transfer learning with 34,209 newly added samples from 35 biomes, including eight novel ones. The transfer neural network model achieved an outstanding average AUROC of 0.989, successfully predicting biome information for newly introduced samples annotated as "Mixed biome." The results of Meta-Sorter are indeed impressive, achieving an overall accuracy rate of 96.7% in classifying samples among the 16,507 lacking detailed biome annotations. This groundbreaking breakthrough effectively resolves the issue of cascading errors and opens up exciting new possibilities for knowledge discovery across various scientific disciplines, particularly in environmental research. Moreover, Meta-Sorter's success extends to refining the biome annotation for under-annotated and mis-annotated samples. Its intelligent and automatic assignment of precise classifications to ambiguous samples provides valuable insights beyond the original literature, while the differentiation of samples into specific environmental categories enhances the reliability and validity of research conclusions.


  • Neural network and transfer learning can improve the biome labeling of microbial communities.
  • Meta-Sorter can consistently classify up to 96.65% of the “Mixed biome” in the EBI-MGnify database.
  • Meta-Sorter prevents the cascading accumulation of inaccurate labels in environmental scientific research.
  • Meta-Sorter enables intelligent and automatic knowledge discovery from microbial communities.

With the ongoing development of standardized protocols for data submission and incorporation of additional meta-data information, Meta-Sorter is set to revolutionize the way researchers analyze and interpret microbial community samples. Ultimately, it will lead to more accurate and insightful discoveries in the realm of microbiome research and beyond.





Funding information

The National Natural Science Foundation of China (32071465, 31871334, and 31671374), The China Ministry of Science and Technology's National Key R&D Program (2018YFC0910502).

About Environmental Science and Ecotechnology

Environmental Science and Ecotechnology (ISSN 2666-4984) is an international, peer-reviewed, and open-access journal published by Elsevier. The journal publishes significant views and research across the full spectrum of ecology and environmental sciences, such as climate change, sustainability, biodiversity conservation, environment & health, green catalysis/processing for pollution control, and AI-driven environmental engineering. The latest impact factor of ESE is 12.6, according to the Journal Citation ReportTM 2022.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.