Predicting variable gene content in Escherichia coli using conserved genes

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Machine Learning in Biology

Background:

• Predicting protein-encoding gene content is crucial for analyzing incomplete or assembled genomes.
• Accurate gene content prediction supports various bioinformatics tasks, including genome quality assessment and risk evaluation.

Purpose of the Study:

• To develop and validate machine learning classifiers for predicting variable gene content in *Escherichia coli* genomes.
• To demonstrate the utility of using nucleotide k-mers from conserved genes as features for gene content prediction.

Main Methods:

• Built 3,259 extreme gradient boosting classifiers to predict the presence/absence of protein families in *E. coli* genomes.
• Utilized nucleotide k-mers from 100 conserved genes as input features.
• Defined orthologs using protein families and focused on genes present in 10%-90% of genomes.

Main Results:

• Achieved a high average macro F1 score of 0.944 across all classifiers.
• Demonstrated accurate prediction of poorly annotated proteins (F1 = 0.902) and genes related to horizontal gene transfer (F1s ranging from 0.824 to 0.895).
• Validated model performance on a diverse holdout set of *E. coli* genomes from environmental sources (average F1 score of 0.880).

Conclusions:

• A robust framework for predicting variable gene content using limited sequence data has been established.
• The developed models are accurate, stable across different *E. coli* strains, and extensible to diverse genomic datasets.
• This approach offers a valuable strategy for enhancing genome analysis and risk assessment in microbial genomics.