Paper Information | SBC BOOKS

Fill in your paper information

Language (*)

Title (*)

Keywords (*)

Abstract (*)

Bacteria can live in a wide range of environmental conditions, from the human gut to boiling water deep in the ocean. Most bacterial species have not been cultivated in the lab. Creating the optimal conditions for some species to thrive in a lab setting may require expensive equipment, expensive materials or several tests with varying environmental conditions. This consumes time and/or resources. Nowadays, due to advances in DNA sequencing technology a bacterial DNA sequence is easily and cheaply obtained. Therefore a tool that can predict bacterial phenotypes from bacterial genomes should be of great help in the search of adequate conditions for bacterial cultivation. In addition, there is a growing number of bacterial Metagenome-Assembled Genomes (MAGs) for which we only have the genome sequence; the ability to predict phenotypes for MAGs would be helpful in understanding the role the corresponding bacteria have in their respective environments. Current tools based on machine learning techniques that can predict bacterial phenotypes based on bacterial genomes (Feldbauer et al., 2020; Weimann et al., 2016) have limited scope. They can predict phenotypic classifications (e.g. if a bacteria is Gram negative or Gram positive, or if it is thermophilic or psychrophilic) but cannot predict numerical phenotypes (e.g. range of pH of the environment that a bacteria can withstand) due to limitations of the machine learning technique chosen. Here we present PHEMALE-Bacteria (PHEnotype MAchine LEarning - Bacteria), a tool capable of predicting both classification and numerical bacterial phenotypes based on the annotated genome sequence. The tool implements multiple machine learning techniques, ranging from linear regression to deep learning. To train the machine learning models used in the tool the phenotypic information was obtained from Madin et al. (2020), which includes data on more than 15000 species of bacteria and archaea, and 4576 bacterial genomes were obtained from the NCBI database (Benson et al., 2012). Preliminary tests achieved precision higher than 97% of F1-macro for sporulation (whether the species sporulates or not) and higher than 0.80 of R² for prediction of pH ranges that the species can withstand. Based on these promising results we plan to include additional phenotypes for prediction. This work was funded in part by a CAPES fellowship to BKVI.

Pages (*)

File Link

English Information

Title

Keywords

Abstract

(*) To change the order drag the item to the new position.

Authors

#	Name
1	Bruno Iha(brunoiha@ime.usp.br,)
2	Carlos Piroupo(morais@iq.usp.br)
3	João Setubal(setubal@iq.usp.br)

(*) To change the order drag the item to the new position.

Reference

#	Reference
1	Benson, D. A., Cavanaugh, M., Clark, K. et al. (2012). GenBank. Nucleic Acids Research, 41 (D1) Feldbauer, R., Hyden, P., Lüftinger, L. (2020). Phenotrex. https://github.com/univieCUBE/phenotrex, July 21st, 2022 Madin, J.S., Nielsen, D.A., Brbic, M. et al. (2020). A synthesis of bacterial and archaeal phenotypic trait data. Scientific Data 7, Article 170 Weimann, A., Mooren, K., Frank, J. et al. (2016). From Genomes to Phenotypes: Traitar, the Microbial Trait Analyzer. mSystems, Nov-Dec 1(6)

Paper Registration

Fill in your paper information

English Information

Authors

Reference