Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data

Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data Nicole M. Davis Diana M. Proctor Susan Holmes David A. Relman Benjamin J. Callahan

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data 2018 Microbiome 10.1186/s40168-018-0605-2 davis2018-simple-statistical-identification davis2018-simple-statistical-identification.txt, the project's staged copy 442b9883d0d89ffc6786e9f80ea4eadb90a0e261 cc-by biology

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them contaminant DNA in marker-gene and metagenomic sequencing - sequences that are in the data and not in the sample - and how to identify them without knowing in advance what they are Quotation withheld because this record is not cleared for redistribution. whether contaminant sequences can be separated from true community members de novo, using only statistical patterns already present in a standard sequencing run Quotation withheld because this record is not cleared for redistribution. decontam: two independent classifiers built on two reproduced signatures of contamination - inverse correlation with DNA concentration, and higher prevalence in negative controls - released as an R package Quotation withheld because this record is not cleared for redistribution. method Quotation withheld because this record is not cleared for redistribution. contamination that survives good laboratory practice and is invisible to the quality control usually used to detect it, biasing hardest exactly where biomass is lowest and the claims are most contested Quotation withheld because this record is not cleared for redistribution. system-or-tool-building system-or-tool-building Quotation withheld because this record is not cleared for redistribution. contaminant DNA enters at roughly constant absolute concentration while sample DNA varies, so as a proportion of the total the contaminant's frequency falls as sample concentration rises - a slope of -1 in log space against a slope of 0 inverse frequency signature Quotation withheld because this record is not cleared for redistribution. with no competing sample DNA, a negative control's sequencing capacity is available to whatever contaminant is present, so contaminants are detected in a larger fraction of controls than of true samples prevalence signature in negative controls Quotation withheld because this record is not cleared for redistribution. because the frequency method reads low sample concentration as evidence of contamination, any real sequence that is genuinely commoner in low-biomass samples presents with the contaminant signature - so mixing environments manufactures false positives out of true biology biomass-correlated confounding Quotation withheld because this record is not cleared for redistribution. decontam is not designed for cross-contamination, which violates its founding assumption that contaminants and true members are distinct populations Quotation withheld because this record is not cleared for redistribution. the frequency method's approximations break down in exactly the low-biomass regime where contamination matters most - when contaminants are a large fraction of reads Quotation withheld because this record is not cleared for redistribution. applying it across environments of systematically different biomass can classify real sequences as contaminants, because low concentration is the contaminant signature Quotation withheld because this record is not cleared for redistribution. it depends on patterns across samples, so it has low sensitivity for contaminants present in very few samples - the tail it cannot see Quotation withheld because this record is not cleared for redistribution. the default classification threshold is not recommended as-is: the authors direct users to read a threshold off their own score histogram instead Quotation withheld because this record is not cleared for redistribution. the method is less effective when groups of samples carry systematically different contaminant patterns, which is why batches must be declared by the user rather than detected Quotation withheld because this record is not cleared for redistribution. model comparison by ratio of sums-of-squared-residuals, carried from linear regression into a per-sequence classifier - and deliberately stripped of the inferential guarantee it carries in its home field Quotation withheld because this record is not cleared for redistribution. a two-component mixture model, carried from statistical modelling to represent total DNA as contaminant plus sample Quotation withheld because this record is not cleared for redistribution. the field's standard check that a negative control shows no band on an agarose gel, taken as evidence that contamination is absent eliminates: contamination in the sequencing library - and the paper's own sentence says this check does NOT eliminate it, because band-less samples still generate non-negligible reads · outcome: absent-stated-about-other Quotation withheld because this record is not cleared for redistribution. removing every sequence that appears in a negative control eliminates: contaminant sequences - and it OVER-removes, because cross-contamination puts abundant TRUE sequences into the negative controls · outcome: absent-stated-about-other Quotation withheld because this record is not cleared for redistribution. discarding sequences below an ad hoc relative abundance threshold eliminates: abundant contaminants, which it fails to remove, while removing rare features that are truly present in the sample · outcome: absent-stated-about-other Quotation withheld because this record is not cleared for redistribution. excluding taxa on a blacklist of previously reported contaminants eliminates: previously reported contaminants - leaving study-specific contaminants in place and removing true sequences · outcome: absent-stated-about-other Quotation withheld because this record is not cleared for redistribution. cross-checking taxonomic assignments against a second controlled vocabulary, HOMD against SILVA, at the genus level eliminates: a contaminant call that is an artefact of one reference vocabulary's genus boundaries · outcome: partial Quotation withheld because this record is not cleared for redistribution. an explicit 'ambiguous' class for genera that neither reference database resolves, instead of forcing every genus into contaminant or oral eliminates: a genus with no evidence either way being forced into 'contaminant' or 'oral' and counted as if it had been decided · outcome: held Quotation withheld because this record is not cleared for redistribution. turning the method on a published low-biomass claim - the placenta microbiome - where the expected answer was already contested eliminates: a decontamination method validated only on data where the right answer was chosen by the people validating it · outcome: held Quotation withheld because this record is not cleared for redistribution. contaminants contaminants Quotation withheld because this record is not cleared for redistribution. External contamination External contamination Quotation withheld because this record is not cleared for redistribution. cross-contamination cross-contamination Quotation withheld because this record is not cleared for redistribution. score statistic score statistic Quotation withheld because this record is not cleared for redistribution. p value p value Quotation withheld because this record is not cleared for redistribution. classification threshold classification threshold Quotation withheld because this record is not cleared for redistribution. negative controls negative controls Quotation withheld because this record is not cleared for redistribution. gel-based quality control gel-based quality control Quotation withheld because this record is not cleared for redistribution. relative abundance thresholds relative abundance thresholds Quotation withheld because this record is not cleared for redistribution. true community members true community members Quotation withheld because this record is not cleared for redistribution. low-biomass low-biomass Quotation withheld because this record is not cleared for redistribution. SILVA SILVA Quotation withheld because this record is not cleared for redistribution. HOMD HOMD Quotation withheld because this record is not cleared for redistribution. processing batches processing batches Quotation withheld because this record is not cleared for redistribution. placenta microbiome placenta microbiome Quotation withheld because this record is not cleared for redistribution. score statistic P - AS COMPUTED: set equal to a p-value from the chi-square or Fisher's exact test, and so arithmetically indistinguishable from one Quotation withheld because this record is not cleared for redistribution. score statistic P - AS LICENSED: a classification score only, explicitly carrying no guarantee on the type 1 error rate. The paper states the difference rather than leaving it to the reader Quotation withheld because this record is not cleared for redistribution. accuracy of decontam - PER SEQUENCE FEATURE: specificity and sensitivity described as sometimes moderate Quotation withheld because this record is not cleared for redistribution. accuracy of decontam - PER READ: described as reaching exceptional levels, on the same classifications of the same data. The unit of account decides which number is reported Quotation withheld because this record is not cleared for redistribution. critiques → gel-band quality control as evidence that a sequencing run is free of contamination Quotation withheld because this record is not cleared for redistribution. critiques → ad hoc in silico decontamination - abundance thresholds, negative-control subtraction and contaminant blacklists Quotation withheld because this record is not cleared for redistribution. builds-on → prior reports of the two contaminant signatures: inverse correlation with DNA concentration, and presence in negative controls Quotation withheld because this record is not cleared for redistribution. confirms → the conclusion that marker-gene data from placenta biopsies do not support an indigenous placenta microbiome Quotation withheld because this record is not cleared for redistribution. score statistic —distinct-from→ p value score statistic p value differs in: whether it licenses a claim about the type 1 error rate · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): whether it licenses a claim about the type 1 error rate · the relation stated: computed the same way and carrying none of the same guarantees Quotation withheld because this record is not cleared for redistribution. contaminants —distinct-from→ true community members contaminants true community members differs in: whether the sequence originated in the sampled environment · the relation itself is separately evidenced · strength of the relation: 0.9 · differs in (stated by the tagger, no quotation): whether the sequence originated in the sampled environment · the relation stated: assumed to be disjoint populations - the assumption the whole method rests on Quotation withheld because this record is not cleared for redistribution. cross-contamination —distinct-from→ External contamination cross-contamination External contamination differs in: whether the contaminating DNA came from outside the study or from another sample in it · the relation itself is separately evidenced · strength of the relation: 0.9 · differs in (stated by the tagger, no quotation): whether the contaminating DNA came from outside the study or from another sample in it · the relation stated: different sources, and different statistical signatures in the data Quotation withheld because this record is not cleared for redistribution. the score decontam reports is computed as a tail probability and is explicitly not a p value - it carries no type 1 error guarantee the score decontam reports is computed as a tail probability and is explicitly not a p value - it carries no type 1 error guarantee Quotation withheld because this record is not cleared for redistribution. the same classifications are 'sometimes moderate' per sequence feature and 'exceptional' per read - the unit of account decides the number the same classifications are 'sometimes moderate' per sequence feature and 'exceptional' per read - the unit of account decides the number Quotation withheld because this record is not cleared for redistribution. the field's usual contamination control - no band on a gel - does not work, because band-less samples still generate substantial reads the field's usual contamination control - no band on a gel - does not work, because band-less samples still generate substantial reads Quotation withheld because this record is not cleared for redistribution. two curated taxonomic databases disagreed on 80 sequences and the disagreement had to be settled by a third method two curated taxonomic databases disagreed on 80 sequences and the disagreement had to be settled by a third method Quotation withheld because this record is not cleared for redistribution. Frequency-based contaminant identification is not recommended for extremely low-biomass samples, where its approximations break down. Frequency-based contaminant identification is not recommended for extremely low-biomass samples, where its approximations break down. Quotation withheld because this record is not cleared for redistribution. DNA sequences not truly present in the sampled community - a definition by provenance, so nothing in the sequence itself can settle it contaminants Quotation withheld because this record is not cleared for redistribution. contributed from outside the samples being measured; distinguished from internal or cross-contamination, which arises when samples mix with each other External contamination Quotation withheld because this record is not cleared for redistribution. a transformation taking the ratio of sums-of-squared residuals and the number of observations, and returning a better classification statistic than the ratio alone - explicitly not a p value score statistic Quotation withheld because this record is not cleared for redistribution. the default classification threshold, which the authors recommend against using unexamined P* = 0.1 Quotation withheld because this record is not cleared for redistribution. the threshold at which classification reduces to 'is the contaminant model the better fit' P* = 0.5 Quotation withheld because this record is not cleared for redistribution. sequences on which the two reference databases disagreed, resolved by BLAST 80 Quotation withheld because this record is not cleared for redistribution. general linear F test general linear F test Quotation withheld because this record is not cleared for redistribution. type 1 error rate type 1 error rate Quotation withheld because this record is not cleared for redistribution. chi-square chi-square Quotation withheld because this record is not cleared for redistribution. Fisher’s exact tests Fisher’s exact tests Quotation withheld because this record is not cleared for redistribution. specificity specificity Quotation withheld because this record is not cleared for redistribution. sensitivity sensitivity Quotation withheld because this record is not cleared for redistribution. sample DNA concentration sample DNA concentration Quotation withheld because this record is not cleared for redistribution. prevalence prevalence Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1186/s40168-018-0605-2.