Input file formats

Our application accepts files in.txt,.tsv, or.csv format. Users can directly upload Level 7 Qiime2 results generated using Greengenes or Silva. Additionally, it supports MEGAN data from whole metagenome sequences (ensure to remove the metadata column if included in the level7.csv file from Qiime2). For Qiime2 input files, the first column serves as an index containing sample IDs, while the second to Nth columns represent taxonomy (S2A Fig). For MEGAN input files, the first seven columns (Level_1 to Level_7) are followed by sample names (S2B Fig). If users wish to upload their files, the first seven columns should contain Kingdom, Phylum, Class, Order, Family, Genus, and Species, followed by sample names (S2C Fig). Metadata files must have two columns. The first column should list sample IDs that match those in the count data input, while the second column, Condition, indicates a user-defined categorical variable, such as “case” and “control” (for two or more groups) (S2D Fig). Users can also refer to our example count data and metadata files available on the tool’s upload page (S3 Fig) or example datasets from our GitHub repository (https://github.com/GudaLab/MetaDAVis).

Guidelines

Once the Input files are uploaded, each of the six modues in MetaDAVis can be used independently in any order. This tool was tested in Linux (RedHat and Ubuntu) and Windows 10 and 11. A user’s manual can be accessed at https://www.gudalab-rtools.net/MetaDAVis/manual/MetaDAVis_manual.pdf to aid in the installation and usage of the application. Summary tables were developed using the DataTables (DT) package to display results in up to 100 rows, while the entire tables can be downloaded as.csv files. Similarly, graphical plots were downloaded in multiple formats using the downloadHandler function from shiny packages. We implemented custom color options using the RColorBrewer package, allowing users to download the figures in the same color format coding format in all the modules except for correlation analysis and MaAsLin3 results. We also provided example data (‘Example Data (To test our tool)’ under the ‘Select Input Format’ section. This feature helps demonstrate our tool’s performance, offering reassurance to users

Data summary and distribution analysis

MetaDAVis accepts the output formats of Qiime2 (generated using Greengenes or Silva) or MEGAN (.csv &.tsv) as input to generate relative abundance and taxonomic distribution plots as per the user selected options (Fig 1A). Relative abundance plots can be generated at seven hierarchical taxonomic levels (Kingdom, Phylum, Class, Order, Family, Genus, and Species) and results can be visualized in multiple plots and tables. For example, a distribution box plot for individual samples and their comparison groups was shown in S4A–S4C Fig.

Diversity analysis

Alpha diversity is calculated using read count or relative abundance data within a sample and compared between groups. We have implemented seven different methods including Observed, Chao1, ACE, Shannon, Simpson, Inverse Simpson, and Fisher from the phyloseq package [18] for α-diversity calculation and the results can be visualized as box or violin plots (Fig 1B) with their summary table (S5A–S5D Fig). Users can also perform the Wilcoxon test to display the statistical significance (p-value) using the microbiomeutilities package [25].

Beta diversity was calculated using phyloseq and vegan [13] packages (Fig 1B). We have incorporated the adonis2 function and defined the parameters such as diversity methods, number of permutations, square root of dissimilarities. Users can choose any one of the methods (bray-curtis, jaccard, manhattan, euclidean, canberra, kulczynski, gower, altGower, morisita, horn, clark, mountford, raup, binomial, chao, cao, mahalanobis, chisq, chord, hellinger, aitchison, and robust.aitchison) for beta diversity calculation using the integrated distance matrices. In addition, users can select any machine learning algorithm including PCoA, NMDS, DCA, RDA, and MDS for assessing the between-sample microbial diversity (S5E–S5I Fig).

Dimension reduction

A critical step in any data analysis is visualizing and summarizing highly variable data in a lower-dimensional space. We implement two and three commonly used dimensionality reduction techniques (Fig 1C) including principal components analysis (PCA) in 2D and 3D with coord_equal(ratio =  1) to get the consistent scale [26], t-distributed stochastic neighbor embedding (t-SNE) [27], and uniform manifold approximation and projection (UMAP) [28]. PCA is a linear dimensionality reduction method where the first three axes explain maximum amount of variation. In contrast, t-SNE and UMAP are non-linear methods for mapping data to a lower-dimensional embedding. We have incorporated six methods from the scater package [29] to plot the t-SNE and UMAP: counts, rclr, hellinger, pa, rank, and relabundance to plot the dimension reduction. The plotted dimension reduction values were provided in separate tables in (S6A–S6L Fig).

Correlation analysis

We implemented both taxon-based and sample-based correlations using GGally, which is an extension to [30] using ggcorr function to call Pearson, Kendall, and Spearman methods (Fig 1D). Users can check the correlation for each condition separately or select multiple options together using the dropdown menu. Similarly, sample-based correlations can be calculated separately for each group of samples under specific conditions or combined across conditions. Correlation plots and summary matrices can be generated by the user with their method of choice (S7A–S7C Fig).

Generating heatmaps

To visualize the relative abundance diversity among the samples, we have implemented a heatmap using ComplexHeatmap [31] and scales [32], with multiple options to display or hide the row and column names and cladograms, clustering methods for rows and columns using options such as single, complete, average (UPGMA), mcquitty (WPGMA), median (WPGMC), and centroid (UPGMC), normalization methods, such as scale, minmax, log, row normalization, column normalization, and none (Fig 1E; S8A and S8B Fig).

Differential abundance analysis

For pair-wise comparison, the generalized linear model-based methods, including DESeq2 [33] and edgeR [34], Two-sample t-test, Wilcoxon Rank Sum test, metagenomeSeq [35], limma-voom [36], Linear Discriminant Analysis Effect Size (LEfSe) lefser [37] and (Microbiome Multivariable Association with Linear Models) MaAsLin3 [38] were used to identify the taxa with different abundances in two different groups. We converted the raw count value to relative frequency using the formula (Relative Frequency =  (Subgroup frequency/ Total frequency) * 100)) for the Wilcoxon Rank Sum test (wilcox.test) and t-test (t.test) statistical analyses. However, metagenomeSeq, DESeq2, Limma-Voom, edgeR lefser and MaAsLin3 have in-built algorithms to find statistically significant biomarkers. For multiple testing, biomarker candidates can be filtered using user-specified p-value or false discovery rate (FDR, q-value) from the Benjamini-Hochberg procedure [39]. Users have the flexibility to adjust the FDR or p-value based on their needs (default is <  0.05). Results can be downloaded either as the grouped or individual box plot for each taxon, or as volcano plots or heatmaps of significantly identified taxa (Fig 1F; S9A–S9F Fig) with summary tables (S2 Table). MaAsLin3 generates multiple tables and figures, and we provide these result files in a compressed zip format for ease of access.

For analyses involving multiple-group comparisons such as control, case 1 and case 2, we implemented the Kruskal-Wallis (kruskal.test) and ANOVA (Analysis of variance) to identify differentially abundant taxonomic markers. In addition, post-hoc test that calculates p-values for pairwise comparisons among the members of the group was implemented using dunn.test package in the Kruskal-Wallis test. Likewise, TukeyHSD was used under ANOVA testing. We have applied the Benjamini-Hochberg FDR or p-value and the post hoc test. These results can be downloaded similar to those from the two-group comparisons.

MetaDAVis provides a graphical user interface through R/Shiny, which can be used even by those without prior programming knowledge. The tools such as vegan, Mothur, MicrobiomeR, Microbiomehelper, and Qiime2 are only command-line interfaces, which limits their usage without prior programming experience. Furthermore, several tools (as presented in Table 1) impose the burden of importing/exporting additional packages, which also requires programming skills. MetaDAVis can be installed locally for standalone use or accessed via an user-friendly web interface to analyze both 16S rRNA and WGS data (Table 1). Hence, it offers more flexibility for use by both seasoned programmers and non-programmers. MataDAVis application is embedded with rich sets of options in each module to choose a variety of methods and perform highly customizable analyses for microbiome sequencing data. For example, it allows users to analyze data at seven different multiple taxonomic levels, provides multiple options for data normalization and distribution analysis, facilitates visualization of data using PCA, t-SNE, UMAP with multiple methods within each approach, and similarly, offers multiple methods to carry out differential abundance analysis, and supports differential analysis and visualization for pairwise and group comparisons. Each of the six modules in MetaDAVis can be used in no particular order using outputs from other methods as inputs, which adds a lot flexibility for users to build and carryout customizable data analysis pipelines. The primary advantage of using MetaDAVis over existing methods is the ease of accessing many independent statistical and machine-learning tools all in one platform, seamlessely, to carryout highly specialized and refined microbiome data analyses using 16S rRNA or WGS datasets. Another advantage is its rich set of visualization tools and graphical outputs. Each module generates publishing quality plots and summary tables, where the images can be downloaded in seven different formats and the tables are downloaded as.CSV files for further use. MetaDAVis is highly flexible for customization of data pipelines and it can be broadly used without any programming background. We believe that MetaDAVis tool is a unique and highly versatile platform that broadly supports microbiome research.