AI Tool Developed by Jerusalem Scientists Identifies Soil Bacteria DNA
Translated & summarized from NEWSru Israel by baba
Researchers at the Hebrew University of Jerusalem have created an AI model called BBERT to identify bacterial DNA in soil, aiding in soil fertility analysis. The model achieves around 90% accuracy in distinguishing bacterial DNA and significantly speeds up genome assembly. BBERT helps classify microbial "dark matter" and offers potential for improved ecological and evolutionary studies.
The story in 5 lines · by baba
- Hebrew University of Jerusalem scientists created an AI model, BBERT, to identify bacterial DNA in soil.
- The AI tool helps analyze soil fertility by identifying microbial communities.
- BBERT distinguishes bacterial DNA from other organisms with approximately 90% accuracy.
- The model reduces genome assembly time from 66 to 25 hours in simulated tests.
- BBERT aids in classifying microbial "dark matter" and improving ecological analysis.
Scientists at the Hebrew University of Jerusalem have developed an artificial intelligence model, named BBERT, designed to identify bacterial DNA fragments in soil samples. This innovation aims to improve the analysis of soil fertility by better understanding the microbial communities present. Soil samples contain DNA from numerous organisms, which, when sequenced, are broken into millions of short fragments from different species. Identifying which fragments belong to bacteria is challenging, especially since many soil microbes are not yet cataloged in reference databases, leading to unclassified or misidentified DNA, often referred to as microbial "dark matter."
The BBERT model functions similarly to a language model but is trained on bacterial DNA sequences instead of text. It can determine if a DNA fragment is bacterial and if it encodes a protein without needing to search external databases. In tests, BBERT accurately distinguished bacterial DNA from other organisms with approximately 90% accuracy and identified protein-coding fragments with 88% accuracy. The model was also used to classify many previously unidentified fragments, revealing that the "dark matter" contains diverse non-bacterial DNA sources beyond just unknown bacteria.
This AI tool significantly speeds up genome assembly, reducing processing time from 66 hours to 25 hours in simulated mixtures compared to standard methods, while also decreasing errors. The model operates locally on a laptop with a suitable graphics card. While BBERT can sometimes confuse bacterial DNA with chloroplast DNA (which has bacterial origins) and occasionally struggles to differentiate bacteria from archaea, the researchers state it "opens access to large metagenomic archives for more accurate ecological and evolutionary analysis." This advancement can help determine soil fertility and suggest methods for improvement.