Comparing genome versus proteome-based identification of clinical bacterial isolates

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Whole-genome sequencing (WGS) is increasingly vital for analyzing bacterial pathogens in clinical settings.
• Current WGS identification tools often rely on pre-existing taxonomic data, potentially introducing bias.
• Accurate and efficient bacterial identification is crucial for diagnosing and managing infectious diseases.

Purpose of the Study:

• To evaluate and compare the performance of five computational tools (CLARK, Kaiju, Kraken, DIAMOND/MEGAN, TUIT) for bacterial identification using WGS data.
• To establish a reliable 'gold standard' for taxonomic classification by comparing expert-driven identification with MALDI-TOF MS.
• To assess the accuracy, F-measure, and runtime of these tools on diverse clinical bacterial isolates.

Main Methods:

• Newly sequenced 846 clinical gram-negative bacterial isolates and 200 *Staphylococcus aureus* isolates.
• Compared five WGS identification tools: CLARK, Kaiju, Kraken, DIAMOND/MEGAN, and TUIT.
• Validated tool performance against expert-driven taxonomy and matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry (MS) analysis.

Main Results:

• CLARK and Kraken (k=31) achieved the highest species classification accuracy (100% and 99.5% for gram-negative and *S. aureus*, respectively).
• CLARK and Kraken demonstrated superior mean F-measure values (85.5–94.7%) compared to other evaluated tools.
• CLARK, Kaiju, and Kraken exhibited significantly faster runtimes, outperforming others by 30 to 170 fold.

Conclusions:

• Nucleotide-based k-mer tools like CLARK and Kraken provide accurate and rapid taxonomic characterization of bacterial isolates from WGS data.
• WGS-based genotyping presents a promising alternative to MS-based biotyping in clinical diagnostics.
• Complementary data and robust evaluation methods are essential for assessing taxonomic classification tools due to potential database limitations.