Removing contaminants from databases of draft genomes

Abstract

Metagenomic sequencing of patient samples is a very promising method for the diagnosis of human infections. Sequencing has the ability to capture all the DNA or RNA from pathogenic organisms in a human sample. However, complete and accurate characterization of the sequence, including identification of any pathogens, depends on the availability and quality of genomes for comparison. Thousands of genomes are now available, and as these numbers grow, the power of metagenomic sequencing for diagnosis should increase. However, recent studies have exposed the presence of contamination in published genomes, which when used for diagnosis increases the risk of falsely identifying the wrong pathogen. To address this problem, we have developed a bioinformatics system for eliminating contamination as well as low-complexity genomic sequences in the draft genomes of eukaryotic pathogens. We applied this software to identify and remove human, bacterial, archaeal, and viral sequences present in a comprehensive database of all sequenced eukaryotic pathogen genomes. We also removed low-complexity genomic sequences, another source of false positives. Using this pipeline, we have produced a database of "clean" eukaryotic pathogen genomes for use with bioinformatics classification and analysis tools. We demonstrate that when attempting to find eukaryotic pathogens in metagenomic samples, the new database provides better sensitivity than one using the original genomes while offering a dramatic reduction in false positives.

Fig 1. Masking procedure.

A) The original genome is split into 100bp overlapping pseudo-reads. B) The pseudo-reads are then classified using Kraken first against the common contaminating vector sequences and the plant, viral, bacterial, archaeal, human, and mouse RefSeq database. The pseudo-reads are also classified using Kraken against non-human and non-mouse vertebrate RefSeq genomes. C) Bowtie2 is then used to align all pseudo-reads against the human genome. D) All pseudo-reads that were classified in the previous steps are masked out of the original genomes. Any remaining non-masked sequence with less than 100p is also masked. E) Finally, Dustmasker is used to mask additional low-complexity sequences.

Fig 2. Masking results.

Fig 2C provides an overview of sequence lengths for each eukaryotic pathogen genome masked in each step and the sequence lengths of the final cleaned genomes. As low-complexity sequences and vertebrate masked sequences are much smaller compared to the final genome length or human/bacterial/viral/plant/vector sequences, these were additionally plotted in Fig 2A and 2B for each eukaryotic pathogen genome. Low-complexity sequences were masked as a final step as well. Masked sequence lengths are also presented as percentages of the original genome length to show the percent of each genome remaining and the percent masked in each step (Fig 2D). Exact numbers are listed in S2 Table.

Fig 3. Pseudo-read Kraken classifications.

The above plot shows the 20 eukaryotic pathogen genomes with the greatest numbers of pseudo-reads that Kraken identified as matching foreign species when searching against database containing bacteria, viruses, archaea, and a limited set of vertebrate genomes. Vertebrate classifications are grouped by common categories, such as fish, birds, rodents, or primates. Primate and rodent numbers do not include human and mouse, which are counted and shown separately. S3 Table contains pseudo-read classifications for all eukaryotic pathogen genomes.

Fig 4. Human/Mouse classified pseudo-reads.

This plot shows the 20 genomes with the most number of pseudo-reads classified as either human or mouse. Perhaps not surprisingly, the mouse strain of malaria, P. yoelii, contains a substantial number of contaminant reads from mouse. S3 Table contains pseudo-read human and mouse classifications for all eukaryotic pathogen genomes.

Fig 5. Top 10 species identified in corneal samples per database.

The non-human reads from the 20 corneal samples were classified against four different Kraken databases: the original EuPathDB (A), EuPathDB-clean (B), RefSeq EuPathDB (C), and the final MicrobeDB (D). The plot above shows the 10 species with the most classified reads per megabase in a single corneal sample.

Fig 6. Number of classified reads per megabase for five true species/genera compared among four databases across all corneal samples.

The above plot compares the reads per megabase for the true pathogens in the infected samples and also shows the reads per megabase from those pathogens in the remaining corneal samples. The five true species/genera are Acanthamoeba (A), Aspergillus flavus (B), Anncaliia algerae (C), Candida albicans/dubliensis (D), and Fusarium (E) S7 Table lists classified reads per megabase for each species for each database.