Clinical PathoScope: rapid alignment and filtration for accurate pathogen identification in clinical samples using unassembled sequencing data

Abstract

BackgroundThe use of sequencing technologies to investigate the microbiome of a sample can positively impact patient healthcare by providing therapeutic targets for personalized disease treatment. However, these samples contain genomic sequences from various sources that complicate the identification of pathogens.ResultsHere we present Clinical PathoScope, a pipeline to rapidly and accurately remove host contamination, isolate microbial reads, and identify potential disease-causing pathogens. We have accomplished three essential tasks in the development of Clinical PathoScope. First, we developed an optimized framework for pathogen identification using a computational subtraction methodology in concordance with read trimming and ambiguous read reassignment. Second, we have demonstrated the ability of our approach to identify multiple pathogens in a single clinical sample, accurately identify pathogens at the subspecies level, and determine the nearest phylogenetic neighbor of novel or highly mutated pathogens using real clinical sequencing data. Finally, we have shown that Clinical PathoScope outperforms previously published pathogen identification methods with regard to computational speed, sensitivity, and specificity.ConclusionsClinical PathoScope is the only pathogen identification method currently available that can identify multiple pathogens from mixed samples and distinguish between very closely related species and strains in samples with very few reads per pathogen. Furthermore, Clinical PathoScope does not rely on genome assembly and thus can more rapidly complete the analysis of a clinical sample when compared with current assembly-based methods. Clinical PathoScope is freely available at: http://sourceforge.net/projects/pathoscope/.Electronic supplementary materialThe online version of this article (doi:10.1186/1471-2105-15-262) contains supplementary material, which is available to authorized users.

PaPrBaG: A machine learning approach for the detection of novel pathogens from NGS data

“…This concerns the metagenome of a clinical sample consisting of a number of highly-abundant, known non-HPs and a single, underrepresented and unknown HP species (for details see section Methods and Table 3). Note that real clinical samples will often be contaminated with reads from the human host, which may be initially subtracted using specialised tools1851.…”

“…reads mapping to different references. Clinical PathoScope18 is particularly useful to remove large numbers of contaminant reads, e.g. human ones in the context of clinical samples.…”

Single-Molecule Long-Read 16S Sequencing To Characterize the Lung Microbiome from Mechanically Ventilated Patients with Suspected Pneumonia

“…PathoScope analysis was performed by mapping reads against a bacterial 16S rRNA data set derived from "The All-Species Living Tree" Project (LTP), supplemented with human sequences (35,40,41). Bowtie 2 was used to map reads using default settings except "--very-sensitive-local Ϫk 100 --score-min L,20,1.0" parameters (42).…”

“…The raw SMRT reads were processed through the PacBio SMRT Portal pipeline to filter out (i) short reads of Ͻ100, (ii) reads with no insert, (iii) trimming of adapter sequences, and (iv) low-complexity or poor-quality reads. Without prior knowledge of the clinical microbiology results, microbial diversity characterization was performed using multiple tools, including (i) PathoScope (34,35), (ii) RDP naive Bayesian classifier (36), (iii) mothur-based rDnaTools application (37), (iv) SMRT Portal (38) (Pacific Biosciences), and (v) Geneious R7 (39) (Biomatters, NZ) software.…”

Using populations of human and microbial genomes for organism detection in metagenomes

“…Marker-based approaches present an important option for fast estimation of microbial content but only tag a small fraction of the input leaving potentially interesting fragments hidden in a large collection of unlabeled reads (Liu et al 2011;Berendzen et al 2012;Segata et al 2012;Sunagawa et al 2013;Minot et al 2014;Tu et al 2014). Recent efforts have demonstrated substantial progress in labeling all reads by applying ordered searches that attempt to reserve the most computationally expensive analysis for the fewest reads (Nakamura et al 2009;Zhao et al 2013;Byrd et al 2014;Cotten et al 2014;Takeuchi et al 2014). A recent example is SURPI , which maps reads to the human reference genome to subtract host reads prior to search of the GenBank NT database.…”