Joint analysis of multiple metagenomic samples - PubMed (original) (raw)
Joint analysis of multiple metagenomic samples
Yael Baran et al. PLoS Comput Biol. 2012.
Abstract
The availability of metagenomic sequencing data, generated by sequencing DNA pooled from multiple microbes living jointly, has increased sharply in the last few years with developments in sequencing technology. Characterizing the contents of metagenomic samples is a challenging task, which has been extensively attempted by both supervised and unsupervised techniques, each with its own limitations. Common to practically all the methods is the processing of single samples only; when multiple samples are sequenced, each is analyzed separately and the results are combined. In this paper we propose to perform a combined analysis of a set of samples in order to obtain a better characterization of each of the samples, and provide two applications of this principle. First, we use an unsupervised probabilistic mixture model to infer hidden components shared across metagenomic samples. We incorporate the model in a novel framework for studying association of microbial sequence elements with phenotypes, analogous to the genome-wide association studies performed on human genomes: We demonstrate that stratification may result in false discoveries of such associations, and that the components inferred by the model can be used to correct for this stratification. Second, we propose a novel read clustering (also termed "binning") algorithm which operates on multiple samples simultaneously, leveraging on the assumption that the different samples contain the same microbial species, possibly in different proportions. We show that integrating information across multiple samples yields more precise binning on each of the samples. Moreover, for both applications we demonstrate that given a fixed depth of coverage, the average per-sample performance generally increases with the number of sequenced samples as long as the per-sample coverage is high enough.
Conflict of interest statement
The authors have declared that no competing interests exist.
Figures
Figure 1. Crohn's disease status separates with components proportions.
Each marker corresponds to an individual, red for Crohn-free and filled black for Crohn cases. The markers are positioned on the two-dimensional plane defined by the components proportions (there are three components but only two dimensions because the proportions sum to 
). The Crohn-free individual at the bottom part of the figure is a colitis case.
Figure 2. Association between
-mers relative abundances and BMI can be corrected using the components proportions. Quantile-quantile curves comparing the uniform distribution to the distribution of the p-values for association between all 
-mers and BMI within the Danish samples. The uncorrected p-values are highly deflated (black), indicating that the abundance of many 
-mers is correlated with BMI. However, when the components proportions are added the the regression equation (red), the correlation disappears for most 
-mers.
Figure 3. Strategies for
-mer choice under the different models. Error is measured as the sum of squared differences between true and estimated 
matrices. The plot presents, for different error thresholds, the number of runs out of 
which yielded a precision at least as small as the threshold. Using the original model, we extracted from each 100 bp read either the first 
-mer, 9 sparely dispersed 
-mer along it, 24 non-overlapping 
-mers or 96 overlapping 
-mers. Extracting multiple 
-mers can be seen to increase precision considerably. Shifting to the refined model yields an even better precision; since this model is more sensitive to dependencies between 
-mers, extracting only the fewer dispersed 
-mers is preferable over extracting all non-overlapping 
-mers.
Figure 4. PLSA approximates mixture coefficient better than PCA.
PCA and PLSA were performed on a simulated counts matrix 
with 
and different number of per-sample counts. The plot shows the average squared correlation coefficient between the true vectors 
and the three strongest principal components (in the case of PCA) or PLSA estimates 
. For each per-sample counts value 20 experiments were performed, and the plot gives the mean result and the standard error of the mean. The estimates obtained by PLSA show higher correlation with the true mixture proportions.
Figure 5. Simultaneous binning over multiple samples achieves higher precision compared with the equivalent single-sample approach.
MultiBin and AbundanceBin were both run on datasets of increasing complexity. Each dataset is composed of 5 mixtures of the specified number of species. The specified precision is the proportion of reads correctly assigned to a bin, averaged over all species. For MultiBin (red) the curves show average precision over 10 random starts of the clustering algorithm, and the error bars give the standard error of the mean. For AbundanceBin (blue) the curves show the average precision over the 5 samples in the dataset, and the dashed lines give the highest and lowest result of the 5. MultiBin achieves consistently better precision over both read lengths and over all sample complexities. AbundanceBin's performance exhibits high between-sample variability, and also deteriorate more rapidly as the number of species increase.
Figure 6. Increasing the number of samples for a fixed depth of coverage improves both components characterization and binning precision.
Left: A fixed number of 
counts were generated from a model defined by uniformly drawn 
and 
matrices using 
. The value of 
, the number of samples, varied from 1 to 1000, and 100 trials were performed for each value. The highest average precision of 
estimation is obtained for 
. Right: A fixed number of 8,192,000 reads of length 400 bp were sampled from different numbers of samples, each consisting of 15 species in uniformly drawn proportions. The smallest average error over all samples was obtained when 32 samples are sequenced. In both plots the error bars give the standard error of the mean.
References
Publication types
MeSH terms
Substances
LinkOut - more resources
Full Text Sources
Other Literature Sources