Evaluating Bioinformatic Pipeline Performance for Forensic Microbiome Analysis^*,†,‡

Abstract

Microbial communities have potential evidential utility for forensic applications. However, bioinformatic analysis of high‐throughput sequencing data varies widely among laboratories. These differences can potentially affect microbial community composition and downstream analyses. To illustrate the importance of standardizing methodology, we compared analyses of postmortem microbiome samples using several bioinformatic pipelines, varying minimum library size or minimum number of sequences per sample, and sample size. Using the same input sequence data, we found that three open‐source bioinformatic pipelines, MG‐RAST, mothur, and QIIME2, had significant differences in relative abundance, alpha‐diversity, and beta‐diversity, despite the same input data. Increasing minimum library size and sample size increased the number of low‐abundant and infrequent taxa detected. Our results show that bioinformatic pipeline and parameter choice affect results in important ways. Given the growing potential application of forensic microbiology to the criminal justice system, continued research on standardizing computational methodology will be important for downstream applications.

Dysbiosis in the Dead: Human Postmortem Microbiome Beta-Dispersion as an Indicator of Manner and Cause of Death

“…While not the first study to classify M/COD from microbial communities (Pechal et al, 2018;Kaszubinski et al, 2019;Zhang et al, 2019), this is the first study to compare random forest classification and logistic regression performance using beta-dispersion. MOD classification success with microbial community random forest indicators alone (Kaszubinski et al, 2019) were comparable to multinomial logistic regression models built with only beta-dispersion (∼40%). Inclusion of case demographic data improved multinomial logistic regression model, which was consistent with previous random forest regression model accuracy of ears and nose body site communities (>60%) (Zhang et al, 2019).…”

“…Inclusion of case demographic data improved multinomial logistic regression model, which was consistent with previous random forest regression model accuracy of ears and nose body site communities (>60%) ( Zhang et al, 2019 ). A strength of the MLR approach, is that it does not depend on specific indicator postmortem microbial taxa, which can vary across studies ( Pechal et al, 2018 ; Kaszubinski et al, 2019 ). Furthermore, we suggest that microbial community information, either taxon dependent (e.g., indicator taxa) or not (e.g., beta-dispersion), could be an additional piece of evidence in M/COD determination.…”

“…We wanted to determine not only which standardization strategy was optimal, but also how sensitive the standardizations were (minimum library size and sample percentage cut off). While rarefaction has been debated (McMurdie and Holmes, 2014; Weiss et al, 2017), we sought to eliminate bias associated with different library sizes that could inflate differences in beta-dispersion among M/CODs (Kaszubinski et al, 2019). Specifically, library size differences can mask biologically meaningful results especially for unweighted distance matrices (Weiss et al, 2017).…”

“…Sequence reads from postmortem microbiome samples were analyzed with QIIME2 (v2018.11) (Bolyen et al, 2019), following the methodology outlined in Kaszubinski et al (2019). DADA2 v1.8.0 (Callahan et al, 2016) was used for denoising.…”

Cold-water Coral Microbiome and Environmental Microbial Communities in a Remote NE Atlantic Submarine Canyon Setting: Microbial Diversity, Coral Health and Prospects

“…That said, it is di cult to draw regional trends and patterns from hostbacterial composition. Meistertzheim et al 16 25 and sequencing platform and/or bioinformatic software 64,65 . We nd that the cross-taxa phenomenon described by Neulinegr et al 21 to be most likely the case.…”