liupfskygre / mdm Goto Github PK

Networks of the co-occurrence relationships present within a microbial community can detect otherwise overlooked associations between species. Additionally, network metrics can quantitatively determine the most important species of a given environment. While many advances have been made in microbial ecology thanks to a network-based approach, most analyses to date have neglected the role of uncharacterized, unknown species called Microbial Dark Matter (MDM). The presence and dominance of MDM across all non-human biomes suggests that this exclusion is a mistake and that the ecological role of MDM must be understood to gain complete knowledge of all microbial interactions encompassing an environment. We need to better understand why MDM persist and with whom they interact on a global scale. More importantly, we need to evaluate whether certain MDM taxa confer a value to the environment that cannot be provided by other, well-characterized microbes.

Here, we present a network-based approach to determine the ecological relevance of MDM. First, all taxa designated as unassigned, ambiguous, or uncultured are defined as MDM. The proportion and prevalence of MDM taxa is then evaluated and compared to other, well-characterized (Known) taxa. Next, to assess the global impact of MDM on the microbial environment, three networks are constructed - including all taxa (called "Original"), excluding MDM (called "Without Unknown"), and excluding an equal number of random knowns (called "Bootstrap"), for each taxonomic classification level from Phylum to Genus. Network centrality measures degree, betweenness, and closeness are calculated and compared between the three network types, to quantitatively determine if MDM removal significantly impacts network shape and structure. Changes in network metrics are statistically evaluated and visualized as boxplots. Finally, to assess which MDM or known taxa most influence their respective microbial communities, hub score is calculated for all taxa present in the Original networks. All results can be validated by checking for early filtering, correlation metric, and sample biases using a range of sample prevalence thresholds, correlation metric tools, and constructing an OTU abundance heatmap for all samples used.

Before starting the MDM Workflow analysis with mdmnets, make sure to have the following installed in R:

library(data.table)
library(reshape2)
library(phyloseq)
library(plyr)
library(dplyr)
library(SpiecEasi)
library(ggplot2)
library(igraph)
library(gridExtra)
library(ggpubr)
library(ggsignif)
library(gplots)
library(devtools)
library(stats)
library(parallel)
library(RColorBrewer)
library(Hmisc)
library(viridis)
library(Matrix)

Now you are ready to begin the MDM Analysis!

You can install mdmnets with:

• devtools::install_github("ConesaLab/MDM")

To see how to use the mdmnets package, please go here