Performance of Microbiome Sequence Inference Methods in Environments with Varying Biomass

Area of Science:

• Microbiology
• Bioinformatics
• Human Health

Background:

• Microbiome community composition is crucial for human health.
• Low-biomass microbial communities are important but challenging to analyze due to contamination and technical noise.
• The impact of varying biomass on sequence processing methods is understudied.

Purpose of the Study:

• To benchmark six different methods for inferring community composition from 16S rRNA sequence reads.
• To evaluate method performance across samples with varying microbial biomass.
• To assess the ability of methods to distinguish true community signals from technical noise and contamination.

Main Methods:

• Compared two operational taxonomic unit (OTU) clustering algorithms, one entropy-based method, and three amplicon sequence variant (ASV) methods.
• Utilized high-biomass mock communities for baseline performance assessment.
• Benchmarked methods on a dilution series of a mock community to evaluate performance across a range of biomass levels.

Main Results:

• ASV methods demonstrated superior sensitivity and precision in high-biomass samples compared to OTU and entropy-based methods.
• Contamination represented an increasing proportion of inferred communities in low-biomass samples, irrespective of the method.
• ASV methods showed the strongest correlation between inferred contaminants and sample biomass, indicating better characterization of both community and contaminants.

Conclusions:

• No single inference method can perfectly separate true sequences from contaminants.
• ASV methods provide a more accurate characterization of microbial communities and associated contaminants, particularly in low-biomass samples.
• The improved accuracy of ASV methods can lead to more robust computational identification of contaminants in challenging low-biomass environments.