deletions | additions
diff --git a/Methods.tex b/Methods.tex
index 54b65bf..aafed24 100644
--- a/Methods.tex
+++ b/Methods.tex
...
Roche 454 FLX system using titanium chemistry at Selah Genomics (Columbia,
SC).
SSU rRNA gene sequences were initially screened by maximum expected errors at
a specific read length threshold
(17) and \citep{edgar2013}. Reads that had less
than~0.5 expected errors at a length of 250 nt were then aligned to the Silva
reference Reference Alignment as provided in the Mothur software package using the Mothur
NAST aligner
\citep{}(21, 22). Anomalous reads \citep{DeSantis2005,schloss2009}. Reads that did not align to the
expected region of the SSU rRNA gene were
discarded discarded. After expected error and
alignment based quality control, 87\% of original reads remained.
Remaining The remaining
reads were annotated using the “UClust” taxonomic annotation framework in
QIIME (18, 19) with \citep{caparaso2010,edgar2010}. We used 97\% cluster seeds from the Silva
SSU rRNA database
\citep{quast2013} as reference
for taxonomic annotation
(provided at QIIME website)
(20). \citep{quast2013}. Sequences were distributed
into OTUs
using the with a centroid based clustering algorithm (i.e. UPARSE
methodology (17). \citep{edgar2013}). The centroid selection also included robust chimera
screening \citep{edgar2013}. OTU centroids were established at a threshold
of 97\% sequence
identity. identity and non-centroid sequences were mapped back to
centroids. Reads that could not be mapped to an OTU centroid at greater
than or equal to 97\% sequence identity were discarded. For phylogenetic
reconstruction, alignment was performed with SSU-Align
(23, 24). \citep{nawrocki2009,nawrocki2013}. Columns in the alignment that were
aligned with poor confidence
(< ($<$ 95\% of characters had posterior
probability
alignment scores
> $>$ 95\%) were
masked. not considered when building
the phylogenetic tree. FastTree
(25) \citep{price2010} was used with default
parameters to build the phylogeny. NMDS ordination was performed on
weighted Unifrac
(32) \citep{lozupone2005} distances. The Phyloseq
(33) \citep{mcmurdie2013} wrapper for Vegan
(34) \citep{oksanen2015} (both
R packages) was used to compute sample values along NMDS axes. The
'adonis' function in Vegan was used to perform Adonis tests (default
parameters)
(36). \citep{Anderson2001a}.
We used
DESeq2, DESeq2 (R package), an RNA-Seq differential expression statistical
framework
(29), \citep{love2014}, to identify OTUs that were enriched in high
density gradient fractions from $^{13}$C-treatments relative to corresponding
density fractions from control treatments (for review of RNA-Seq differential
expression statistics applied to microbiome OTU count data see (30)). We define
"high density gradient fractions" as gradient fractions whose density falls
between 1.7125
- and 1.755 g ml$^{-1}$. Briefly, DESeq2 includes several features that
enable robust estimates of standard error in addition to reliable ranking of
logarithmic fold change (LFC) in abundance (i.e. gamma-Poisson regression
coefficients) even with low count groups where LFC can often be noisy CITE.
Further, statistical evaluation of LFC can be performed with selected
...
one standard deviation above the mean of all LFC values. P-values were
corrected for multiple comparisons by using the Benjamini and Hochberg (BH)
method (31). Independent filtering was performed on the basis of sparsity prior
to correcting P-values for multiple comparisons. The sparsity
threshold value that
yielded the most P-values less than 0.10 was used for
sparsity independent
filtering. filtering by
sparsity. Briefly, OTUs were eliminated if they failed to appear in at least
XX\% 45\% of high density gradient fractions for a given $^{13}$C and control
treatment pair, these OTUs are unlikely to have sufficient data to allow for
the determination of statistical significance.