aMeta: ancient metagenomics profiling workflow overview. The workflow represents a combination of taxonomic classification + filtering steps with KrakenUniq that allows to establish a list of microbial candidates for further building a MALT database, running LCA-based alignments with MALT against the database, and performing validation + authentication analysis based on the alignments.

Contexts in source publication

Context 1

... aMeta workflow overview is shown in Figure 1. It represents an end-to-end processing and analysis framework implemented in Snakemake [24] that accepts raw data as a set of files, usually belonging to a common project, and outputs a ranked list of detected ancient microbial species together with their abundances for each sample, as well as a number of validation and authentication plots for each identified microorganism in each sample. ...

Context 2

... a microbe is truly detected, the reads should map evenly across the reference genome, see Figure 2B. In contrast, in case of misaligned reads, i.e. when reads originating from species A map to the reference genome of species B, it is common to observe "piles'' of reads aligned to a few conserved regions of the reference genome, which is the case in Figure 2A (see also Supplementary Figure 1 for a real data example, where reads from unknown microbial organisms are forced to map to Yersinia pestis reference genome alone). Therefore, we consider the breadth of coverage information delivered by KrakenUniq to be of crucial importance for robust filtering in our workflow. ...

Context 3

... KrakenUniq and aMeta have higher sensitivity for microbial detection. This conclusion is confirmed by Supplementary Figures 7-12, where the ground truth for the microbial presence-absence per sample is compared against the one reconstructed by aMeta and HOPS. For example, such simulated species as Campylobacter rectus, Fusarium fujikuroi, Methylobacterium bullatum, Micromonas commoda, Micromonospora echinospora and Mycolicibacterium aurum were correctly identified by aMeta but not detected by HOPS in any simulated sample. ...

Context 4

... ran aMeta and HOPS with default settings on the simulated ground truth dataset, and obtained lists of microbial organisms ranked by the scoring system of aMeta and HOPS, where likely present and ancient microbes received higher scores. Visual inspection of the native heatmap output from HOPS demonstrated its poor authentication performance, Supplementary Figure 13. More specifically, a few bacteria such as Rhodopseudomonas palustris, Rhodococcus hoagii, Lactococcus lactis, Brevibacterium aurantiacum, Burkholderia mallei were erroneously reported by HOPS to be ancient as they got the highest scores in several samples, while they were supposed to be modern according to the simulation's design. ...

Context 5

... were, however, correctly ranked low as potential modern contaminants by aMeta. In contrast, the simulated ancient Salmonella enterica genome was ranked low by HOPS due to read misalignment, while it obtained high scores from aMeta correctly indicating its presence and ancient status, see Supplementary Figures 14-15. Overall, we conclude that aMeta has a lower authentication error compared to HOPS, see Supplementary Information S6 for more details. ...

Context 6

... discovered that nearly 22 000 reads mapped uniquely, i.e. with mapping quality MAPQ > 0, Supplementary Figure 3. Since the sample was from a modern infant who unlikely suffered from plague, the mapped reads cannot be used as evidence of Y. pestis presence in the infant's stool sample. Further, visually inspecting the alignments in Integrative Genomics Viewer (IGV) [29] we confirmed that the reads aligned unevenly implying Y. pestis was not a right reference genome for the reads, see Supplementary Figure 1. Assuming that a large fraction of the aligned reads might be of human rather than bacterial origin, and thus misaligned to the Y. pestis reference genome due to the absence of a human reference genome in the reference database, we concatenated the hg38 human reference genome, https://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/, ...

Context 7

... substantial constraint of HOPS is that it does not provide information on the breadth and evenness of coverage. As discussed in the main text and shown in Figure 2, this often results in Supplementary Figure 1. ...

Context 8

... belonging to the HOPS toolbox, by default assigns +1 to any microbial hit with non-zero terminal damage. Therefore, even a tiny transition mutation frequency, such as 0.001 at the terminal ends of the reads, counts as the presence of damage, which substantially inflates the authentication error, see Supplementary Figure 13. ...

Context 9

... ground truth abundance matrix was binarized using 1% of all reads as a detection threshold, and a binary (present / absent) heatmap of distribution of microbial organisms across samples is presented in Supplementary Figure 8. To compare the microbial detection accuracy of HOPS and aMeta, we built binary heatmaps based on the microbial abundances reconstructed by HOPS and aMeta, see Supplementary Figures 9 and 10, and compared them with the ground truth in Supplementary Figure 8. For aMeta, default internal filter settings were used to determine the presence / absence of a microbe, while we used a reasonable threshold of 300 reads to binarize the HOPS microbial abundance matrix. ...

Context 10

... summarize the results, we computed the accuracy of microbial detection, i.e., presence or absence irrespective of the ancient status, based on the confusion matrices for aMeta and HOPS, see Supplementary Figure 11. The confusion matrix corresponding to aMeta had a detection accuracy of 86% and was more balanced compared to the HOPS confusion matrix, which had an accuracy of 69% and a very high false-negative rate. ...

Context 11

... /2022 filtering) to 800 reads (very harsh filtering), see Supplementary Figure 12. Within the tested range of read number cutoffs, the accuracy of HOPS microbial detection was consistently lower than that of aMeta and did not show substantial variation. ...

Context 12

... native HOPS scoring system includes only three quality metrics: deamination profile and edit distances for all reads and damaged reads. This not only results in many obvious false-positive discoveries, as shown in Supplementary Figure 13, but also too little variation in scores that is not straightforward to align with aMeta's seven-metric scoring system to compute a ROC-curve. Therefore, for a proper comparison, we had to modify the native HOPS scoring system and add reasonable assessments of depth and evenness of coverage, as well as PMD scores and read length distribution. ...