Normalization and microbial differential abundance strategies depend upon data characteristics.

Weiss S, Xu ZZ, Peddada S, Amir A, Bittinger K, Gonzalez A, Lozupone C, Zaneveld JR, Vázquez-Baeza Y, Birmingham A, Hyde ER, Knight R. · 2017

Read Normalization and microbial differential abundance strategies depend upon data characteristics. by Weiss S, Xu ZZ, Peddada S, Amir A, Bittinger K, Gonzalez A, Lozupone C, Zaneveld JR, Vázquez-Baeza Y, Birmingham A, Hyde ER, Knight R. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5335496.txt.

Introduction

Sophie Weiss and co-authors tested data methods for microbe studies in Microbiome in 2017. Gene reading of a marker called 16S gives counts of microbes in a sample. Those counts are tricky: sample sizes swing over wide ranges, most entries are zero, and shares must sum to one. Standard math tools bend or break on such data. Labs must first balance, or normalize, the counts.

The team ran known balancing methods and group tests on data with varied traits. Most methods grouped samples by true source well when group mixes differed a lot. One older approach, thinning samples to one size, grouped best on some distance measures. But no method won across the board. Which choice is sound hinged on how strong the group gap was and how the data spread. The paper gives a guide for picking methods by data shape.

Catalog record

CREATORWeiss S, Xu ZZ, Peddada S, Amir A, Bittinger K, Gonzalez A, Lozupone C, Zaneveld JR, Vázquez-Baeza Y, Birmingham A, Hyde ER, Knight R.
YEAR2017
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC5335496

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Normalization and microbial differential abundance strategies depend upon data characteristics By Sophie Weiss, Zhenjiang Zech Xu, Shyamal Peddada, Amnon Amir, Kyle Bittinger, Antonio Gonzalez, Catherine Lozupone, Jesse R. Zaneveld, Yoshiki Vázquez-Baeza, Amanda Birmingham, Embriette R. Hyde, Rob Knight Published in: Microbiome DOI: 10.1186/s40168-017-0237-y ABSTRACT Background Data from 16S ribosomal RNA (rRNA) amplicon sequencing present challenges to ecological and statistical interpretation. In particular, library sizes often vary over several ranges of magnitude, and the data contains many zeros. Although we are typically interested in comparing relative abundance of taxa in the ecosystem of two or more groups, we can only measure the taxon relative abundance in specimens obtained from the ecosystems. Because the comparison of taxon relative abundance in the specimen is not equivalent to the comparison of taxon relative abundance in the ecosystems, this presents a special challenge. Second, because the relative abundance of taxa in the specimen (as well as in the ecosystem) sum to 1, these are compositional data. Because the compositional data are constrained by the simplex (sum to 1) and are not unconstrained in the Euclidean space, many standard methods of analysis are not applicable. Here, we evaluate how these challenges impact the performance of existing normalization methods and differential abundance analyses. Results Effects on normalization: Most normalization methods enable successful clustering of samples according to biological origin when the groups differ substantially in their overall microbial composition. Rarefying more clearly clusters samples according to biological origin than other normalization techniques do for ordination metrics based on presence or absence. Alternate normalization measures are potentially vulnerable to artifacts due to library size. Effects on differential abundance testing: We build on a previous work to evaluate seven proposed statistical methods using rarefied as well as raw data. Our simulation studies suggest that the false discovery rates of many differential abundance-testing methods are not increased by rarefying itself, although of course rarefying results in a loss of sensitivity due to elimination of a portion of available data. For groups with large (~10×) differences in the average library size, rarefying lowers the false discovery rate. DESeq2, without addition of a constant, increased sensitivity on smaller datasets (<20 samples per group) but tends towards a higher false discovery rate with more samples, very uneven (~10×) library sizes, and/or compositional effects. For drawing inferences regarding taxon abundance in the ecosystem, analysis of composition of microbiomes (ANCOM) is not only very sensitive (for >20 samples per group) but also critically the only method tested that has a good control of false discovery rate. Conclusions These findings guide which normalization and differential abundance techniques to use based on the data characteristics of a given study. Electronic supplementary material The online version of this article (doi:10.1186/s40168-017-0237-y) contains supplementary material, which is available to authorized users. BACKGROUND Although data produced by high-throughput sequencing has been proven extremely useful for understanding microbial communities, the interpretation of these data is complicated by several statistical challenges. Following initial quality control steps to account for errors in the sequencing process, microbial community sequencing data is typically organized into large matrices where the columns represent samples, and the rows contain observed counts of clustered sequences commonly known as operational taxonomic units, or OTUs, that represent bacteria types. These tables are often referred to as OTU tables. Several features of OTU tables can cause erroneous results in downstream analyses if unaddressed. First, the microbial community in each biological sample may be represented by very different numbers of sequences (i.e., library sizes), reflecting differential efficiency of the sequencing process rather than true biological variation. This problem is exacerbated by the observation that the full range of species is rarely saturated, so that more bacterial species are observed with more sequencing (similar trends by sequencing depth hold for discovery of genes in shotgun metagenomic samples [1, 2]). Thus, samples with relatively few sequences may have inflated beta (β, or between sample) diversity, since authentically shared OTUs are erroneously scored as unique to samples with more sequences [3]. Second, most OTU tables are sparse, meaning that they contain a high proportion of zero counts (~90%) [4]. This sparsity implies that the counts of rare OTUs are uncertain, since they are at the limit of sequencing detection ability when there are many sequences per sample (i.e., large library size) and are undetectable when there are few sequences per sample. Third, the total number of reads obtained for a sample does not reflect the absolute number of microbes present, since the sample is just a fraction of the original environment. Since the relative abundances sum to 1 and are non-negative, the relative abundances represent compositional data [5–7]. Compositional data are constrained by the simplex (sum to 1) and are not free floating in the Euclidean space; therefore, standard methods of analysis are not applicable. For example, an increase in abundance of one prevalent bacterial taxon can lead to spurious negative correlations for the abundance of other taxa [8]. Uneven sampling depth, sparsity, and the fact that researchers are interested in drawing inferences on taxon abundance in the ecosystem using the specimen level data represent serious challenges for interpreting data from microbial survey studies. In an attempt to mitigate some of these three challenges and aid in data interpretation, data are often normalized by various computational processes prior to downstream analysis. Normalization is the process of transforming the data in order to enable accurate comparison of statistics from different measurements by eliminating artifactual biases in the original measurements. For example, in microbiome data, biases that reflect no true difference in underlying biology can exist due to variations in sample collection, library preparation, and/or sequencing and can manifest as, e.g., uneven sampling depth and sparsity. After effective normalization, data from different samples can then be compared to each other. Ordination analysis, such as principal coordinate analysis (PCoA) [9], is often then applied to these normalized data to visualize broad trends of how similar or different bacterial populations are in certain sample types, such as healthy vs. sick patients (ordination is a general term for a family of techniques that summarize and project multivariate community data into lower-dimension space). This enables easy visual inspection of sample groupings, driven by sample bacterial content similarity/dissimilarity, and any association with sample metadata. Researchers may further wish to determine, through statistical testing, which specific bacteria are significantly differentially abundant between two ecosystems; this process is known as differential abundance testing. Significant changes in certain bacterial species abundances are linked to inflammatory bowel diseases [10], diarrhea [11], obesity [12–14], HIV [15], diet [16], culture, age, and antibiotic use [17], among many other conditions. However, the reliability of these findings depends upon how much the statistical challenges posed by the underlying community sequence data impact the chosen normalization and differential abundance testing techniques. This paper therefore examines how various normalization and differential abundance testing procedures available in the literature are affected by the challenges inherent in microbiome data. Recent work in this area [18] addresses the performance of parametric normalization and differential abundance testing approaches for microbial ecology studies, but it is primarily focused on estimating proportions. We update and expand those findings using both real and simulated datasets exemplifying the challenges noted above. Normalization

Read the complete text