Page 37 - Read Online

Fabbrini et al. Microbiome Res Rep 2023;2:25 https://dx.doi.org/10.20517/mrr.2023.25 Page 7 of 18

Table 1. List of the most common tools used for microbiome networking
TOOL METHOD FORM SOURCE
Correlation methods
SparCC Log-transformed compositional Pearson correlation -Python https://github.com/dlegor/SparCC.git
-R package
CCLasso Log-ratio transformed abundances in latent variable R package https://github.com/huayingfang/CCLasso.git
model with L1-norm shrinkage method
Canonical correlation Correlation matrices (Pearson, Spearman), custom -Custom R https://igraph.org/
and igraph network weighted coefficient thresholds and multiple testing codes
analysis correction -R package
CoNet Canonical correlation but with plenty of possible Cytoscape https://apps.cytoscape.org/apps/conet
presets in a user-friendly interface app
Graphical models
MInt Poisson-multivariate normal hierarchical model R package https://rdrr.io/cran/MInt/
SPIEC-EASI Neighborhood selection and sparse inverse R package https://github.com/zdk123/SpiecEasi.git
covariance selection
SPRING Semi-parametric rank-based correlation estimation R package https://github.com/GraceYoon/SPRING.git
with Meinshausen and Bühlmann neighborhood
selection and stability-based approach for optimal
tuning
HARMONIES Zero-inflated negative binomial distribution -R package -
-Web tool https://github.com/shuangj00/HARMONIES.git
- https://lce.biohpc.swmed.edu/harmonies/
FlashWeave Local-to-global learning constraint-based causal Julia CLI https://github.com/meringlab/FlashWeave.jl.git
inference framework
MDiNE Bayesian graphical model fit with MCMC methods R package https://github.com/kevinmcgregor/mdine.git
NetCoMi Wrapper of network construction tool with tailored R package https://github.com/stefpeschel/NetCoMi.git
corrections ad adjustments
Multi-omic data
integration
DIABLO Multi-omic integration via correlation and dimension R package http://mixomics.org/mixdiablo/
reduction methods https://doi.org/doi:10.18129/B9.bioc.mixOmics
MiBiOmics Multi-omic integration via correlation and dimension -R package - https://gitlab.univ-nantes.fr/combi-
reduction methods -Web tool ls2n/mibiomics
- https://shiny-bird.univ-
nantes.fr/app/Mibiomics

[44]
Tools implementing this logic to its simplest extent include: (i) MInt , which implements a Poisson-
multivariate normal hierarchical model to learn direct interactions from compositional data; (ii) SPIEC-
EASI , which relies on a two-step inference of the interaction graph from the transformed compositional
[45]
data, namely neighborhood selection and sparse inverse covariance selection; (iii) SPRING , which infers
[46]
graphical models via semi-parametric rank-based correlation estimation with Meinshausen and Bühlmann
neighborhood selection for the identification of sparse conditional dependencies from such estimated
[47]
correlation and a final stability-based approach for optimal tuning of algorithm parameters; and (iv)
HARMONIES , implementing a zero-inflated negative binomial distribution to model the skewness and
[48]
sparsity of microbiome data. All these tools address part of the issues related to the nature of microbiome
data and provide reasonable, sparse, and interpretable networks.

Another tool worth mentioning is FlashWeave, which is based on the local-to-global learning (LGL)
approach, a constraint-based causal inference framework for predicting direct relationships between
variables. Due to the conservative handling of structural zeros, FlashWeave can show reduced statistical
power, hampering smaller datasets. Yet, it exhibits a significant increase in both speed, by several orders of
magnitude, and network quality compared to alternative methods, especially when dealing with
[49]
heterogeneous sequencing data .