Castanet: a pipeline for rapid analysis of targeted multi-pathogen genomic data

Abstract

Motivation: Target enrichment strategies generate genomic data from multiple pathogens in a single process, greatly improving sensitivity over metagenomic sequencing and enabling cost-effective, high-throughput surveillance and clinical applications. However, uptake by research and clinical laboratories is constrained by an absence of computational tools that are specifically designed for the analysis of multi-pathogen enrichment sequence data. Here we present an analysis pipeline, Castanet, for use with multi-pathogen enrichment sequencing data. Castanet is designed to work with short-read data produced by existing targeted enrichment strategies, but can be readily deployed on any BAM file generated by another methodology. Also included are an optional graphical interface and installer script.

Results: In addition to genome reconstruction, Castanet reports method-specific metrics that enable quantification of capture efficiency, estimation of pathogen load, differentiation of low-level positives from contamination, and assessment of sequencing quality. Castanet can be used as a traditional end-to-end pipeline for consensus generation, but its strength lies in the ability to process a flexible, pre-defined set of pathogens of interest directly from multi-pathogen enrichment experiments. In our tests, Castanet consensus sequences were accurate reconstructions of reference sequences, including in instances where multiple strains of the same pathogen were present. Castanet performs effectively on standard computers and can process the entire output of a 96-sample enrichment sequencing run (50M reads) using a single batch process command, in $<$2 h.

Availability and implementation: Source code freely available under GPL-3 license at https://github.com/MultipathogenGenomics/castanet, implemented in Python 3.10 and supported in Ubuntu Linux 22.04. The data underlying this article are available in Europe Nucleotide Archives, at https://www.ebi.ac.uk/ena/browser/view/PRJEB77004.

Figure 1.

Castanet process flow. (Left) Castanet pipeline. Blue boxes are optional stages. (Right) Consensus generator algorithm.

Figure 2.

Castanet output for set A.1, at several dilutions. (a) Read depth graph showing total and deduplicated (unique) reads for HPeV, undiluted, viral load 9.6 × 10⁴ copies per ml. (b) Read depth, HPeV, 1:500 dilution, 5.3 × 10² copies per ml. (c) Log10 deduplicated reads versus Log10 copies per ml for HPeV, HHV4, and HHV5 at four different dilutions (1, 1:10, 1:100, and 1:500), with linear regression line.

Figure 3.

Appearance of presumptive low positives, false positives, and contamination in read depth plots from set A.2. (a) Reads for TTV, with high amplification rate across the region with reads (4.0), across a partial region of the genome. (b) Trypanosoma spp. reads at the ribosomal 18s locus, with no amplification and incomplete coverage. (c) HBV reads with incomplete coverage and significant amplification (MAR 5.82, SD 3.37).

Figure 4.

Comparison of read depth between untargeted and targeted sequences, in a screened pooled human plasma sample (set A.2). (a) Whole human mitochondrial gene, median depth 16 (excluding COX-1 region). (b) Targeted sequence, mitochondrial COX-1, median depth 2109.