DAAMS

Differential abundance analysis for shotgun metagenomics. This repo is inspired by https://github.com/nearinj/Comparison_of_DA_microbiome_methods and uses their code as the starting code. Changes are made here to include multiple variables for some methods and also more suitable for shotgun metagenomics data.

For quick start take a look at analysis.Rmd

Summary

This is a simple wrapper that invokes several methods for differential abundance analysis.

For a single fixed variable the following methods are run:

1. DeSeq
2. Maaslin2
3. EdgeR
4. MetagenomeSeq
5. ANCOM2
6. ANCOM-BC

For two fixed variables the following methods are run:

1. DeSeq
2. Maaslin2
3. ANCOM-BC

For fixed variables and a random variable (like subject in repeated measures) the following methods are run:

1. DeSeq
2. Maaslin2
3. ANCOM-BC

Input

The main function to use is wrapper_daa() and following inputs are required by this function:

1. A comma separated species table is required. See example

2. A metadata file, see example

3. The name of the column in the metadata file that is the variable of interest.

4. Set min_abund: Minimum total abundance of a taxa required to be kept for further analysis

5. Set min_prevelance: Minimum number of samples that a taxa should be present in.

6. Output_dir: Output directory where all the results are saved.

An analysis.Rmd file has the complete workflow that can be used for Metaphlan3 results and for pathway abundace tables from Humann3.

Method details and shortcommings

Most of the methods used here have a lot of flexibility and the functions can take a lot more parameters as input. The table below explains briefly how these methods are used here. Also all output files are found in the output directory specified by the user.

Method	function Name	Output files	Information	What features are used	What is not used
DESeq2	Deseq_fun()	deseq.csv	UserGuide	Works for two variables	Ramdom variables are added to the model like (Fixed + random). No interaction terms are allowed
EdgeR	edgeR_fun()	edgeR.csv	UserGuide	Works for one
Maaslin2	maaslin2_fun()	Maaslin2.csv + Maaslin2 dir	UserGuide	works for two variables + random variables ,CSS normalization is default and no transformation is applied	Not much missing but you can change the normalization method or add transformation
MetagenomeSeq	metagenomeSeq_fun()	Metagenomeseq.csv	UserGuide	works for one variable	It can be extended to multiple variables as per the user guide but looks more difficult to include
ANCOM2	AMCOM2_fun()	ANCOM2.csv	UserGuide	works for one variable + 1 random variable	structural zero feature is not being used right now but it can be use if the ANCOM2fun is called directly
ANCOM-BC	ANCOMBC_fun()	ANCOMBC.csv + ACNOMBC_sample_fracs.csv	UserGuide	works for two variables + 1 random variable	Bias correction is not done but the sample fractions are saved if anyone wants to do it later.

Suitable Normalization methods

For some methods, there is an option to chose a normalization or transformation method. Based on recent publication Pereira et. al 2018 methods that perform the best (Low false postives (FG) and high true positve rates (TP)) are RLE( relative log expression), TMM(trimmed mean of M-values) and CSS (cumulative sum scaling) for larger dataset.

Required R packages

tidyverse v1.3.1

limma v3.50.0

knitr v1.37

edgeR v3.36.0

Maaslin2 v1.8.0

metagenomeSeq v1.36.0

rstatix v0.7.0

phyloseq v1.38.0

DESeq2 v1.34.0

ANCOMBC V1.4.0

DT v0.20