Dirichlet-multinomial modelling outperforms alternatives for analysis of microbiome and other ecological count data

Abstract

Molecular ecology regularly requires the analysis of count data that reflect the relative abundance of features of a composition (e.g., taxa in a community, gene transcripts in a tissue).The sampling process that generates these data can be modeled using the multinomial distribution. Replicate multinomial samples inform the relative abundances of features in an underlying Dirichlet distribution. These distributions together form a hierarchical model for relative abundances among replicates and sampling groups. This type of Dirichletmultinomial modelling (DMM) has been described previously, but its benefits and limitations are largely untested. With simulated data, we quantified the ability of DMM to detect differences in proportions between treatment and control groups, and compared the efficiency of three computational methods to implement DMM-Hamiltonian Monte Carlo (HMC), variational inference (VI), and Gibbs Markov chain Monte Carlo. We report that DMM was better able to detect shifts in relative abundances than analogous analytical tools, while identifying an acceptably low number of false positives. Among methods for implementing DMM, HMC provided the most accurate estimates of relative abundances, and VI was the most computationally efficient. The sensitivity of DMM was exemplified through analysis of previously published data describing lung microbiomes. We report that DMM identified several potentially pathogenic, bacterial taxa as more abundant in the lungs of children who aspirated foreign material during swallowing; these differences went undetected with different statistical approaches. Our results suggest that DMM has strong potential as a statistical method to guide inference in molecular ecology.In many scientific disciplines, data from both manipulative experiments and surveys of nat-2 ural variation are often counts of observations that are assigned to categories. Given some 3 total level of observational effort, the counts of the different features in the sample (e.g., 4 2 taxa or transcripts) reflect the underlying proportions of those features in the sampled com-5 position (e.g., an assemblage of organisms or collection of molecules). In molecular ecology, 6 such sampling can take the form of detecting and counting taxa based on observed DNA 7 sequences (e.g., in molecular barcoding or microbial ecology) or counting the reads assigned 8 to specific transcripts in studies of gene expression , Gloor et al. 2017 Tsilimigras and Fodor 2016). For these applications, sampling effort corresponds to the total 10 number of sequence reads, and the count of reads assigned to a taxon or gene supports infer-11 ence of their true proportion in the composition. Moreover, the total number of reads that 12 can be obtained is constrained by the sequencing instrument, with reads ascribed to samples 13 and features within each sample. Due to this constant sum constraint, compositional data 14 have the important quality that as the relative abundance of one feature in the composition 15 increases, other featur...

The quest for absolute abundance: The use of internal standards for DNA‐based community ecology

“…We note that if extreme sequencing depth is employed, such as what can be obtained through the Illumina NovaSeq platform, it may be possible to use much less ISD and still achieve satisfactory results. We also suggest that a modeling approach to estimate proportions from count data for all sequenced features should allow much lower input of ISD than would estimation of proportions following rarefaction, because accurate estimates of proportions can be modeled given few observations (Harrison, Calder, Shastry, & Buerkle, 2020). Also, we note that if a cellular ISD is used for metabarcoding studies it is wise to consider the CNV of the focal loci when performing concentration calculations prior to ISD addition (see Stämmler et al., 2016).…”

“…The bulk of the variation they observed was assigned to technical causes. We suggest that Bayesian models are an exciting possibility for partitioning variation in sequence data, in part because they make full use of the data and can incorporate hierarchical model structures to share information among all replicates within a sampling group (sensu Fordyce, Gompert, Forister, & Nice, 2011; Harrison et al., 2020). This is in contrast to rarefaction methods, which discard observed data and thus provide potentially misleading information about technical and biological variation among samples (McMurdie & Holmes, 2014).…”

A suite of rare microbes interacts with a dominant, heritable, fungal endophyte to influence plant trait expression

“…We analyzed sequence count data via a hierarchical Bayesian modeling (HBM) framework that provides estimates of proportional relative abundance for each microbial taxon [67,68].…”

Dirichlet‐multinomial modelling outperforms alternatives for analysis of microbiome and other ecological count data

“…All scripts and processed data used for this manuscript are available at https://github.com/JHarrisonEcoEvo/DMM Harrison, Calder, Shastry, & Buerkle, and a snapshot corresponding to the status at publication at Zenodo (10.5281/zenodo.3558682). Data from Duvallet et al () can be downloaded from (https://doi.org/10.5281/zenodo.2678108).…”