Evaluation of methods for detecting human reads in microbial sequencing datasets

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Host-associated microbial sequencing is susceptible to human DNA contamination from investigators or subjects.
• Accurate removal of human DNA is crucial for reliable microbial analysis.

Purpose of the Study:

• To benchmark various computational methods for detecting and removing human DNA contamination from microbial sequencing data.
• To inform best practice guidelines for handling such contamination.

Main Methods:

• Evaluated eight alignment-based and two classification-based human read detection methods.
• Used simulated data with known proportions of bacterial, viral, and human reads.
• Re-analyzed 11,577 public bacterial sequencing datasets for human contamination.

Main Results:

• Most methods detected >99% of human reads, but precision varied.
• Bowtie2 and SNAP showed high precision with minimal misclassification of microbial reads.
• Classification methods (Kraken2, Centrifuge) were effective for viral and longer bacterial reads but could misclassify shorter bacterial reads.
• Human reads, though <0.1%, were non-randomly distributed, often from sex chromosomes.
• Detected significant human contamination in 6% of public datasets, revealing clinically relevant human SNPs.

Conclusions:

• The choice of human DNA removal method impacts accuracy and should consider read length and data type.
• Combined or sequential methods (e.g., Bowtie2 followed by SNAP) offer robust human DNA removal for short bacterial reads.
• Human DNA contamination is prevalent in public microbial datasets and can contain actionable genetic information.