metaboanalyst

MetaboAnalyst

This is a web app with a large suite of tools for various uses. It's based on MetaboAnalystR v2.0, the R package.

Capabilities:

• Statistical analysis^ - Some data processing, e.g. normalization, log transformation (if applicable), for t-tests, ANOVA, volcano plots, k-means, random forest, etc... too many to list.
• Enrichment analysis - Enrichment analysis aims to evaluate whether the observed genes and metabolites in a particular pathway are appearing more/less than expected by random chance within the given dataset. Given an input list of compounds and (optionally) concentrations (input in a text field or uploaded as .csv), compare against a metabolite set of your choice (e.g. blood, urine, CSF, or your own custom metabolite set) to see which compound appears more/less than "usual" compared to the selected metabolite set, in terms of fold-change and p-value.
• Pathway enrichment analysis - Pathway analysis (a.k.a. topology/network analysis) aims to evaluate whether a given gene or metabolite plays an important role in a biological response based on its position within a pathway/network. Uses graph theory concepts like node centrality, in-between-ness, etc. Given an input list of compounds and (optionally) concentrations (input in a text field or uploaded as .csv), map onto a known network (e.g. human KEGG/SMPDB, fruitfly, yeast, etc.), with some topological analysis algorithms that do network clustering. Outputs a heatmap of pathway impact (effect size), with associated p-values. Results can be downloaded in a zip file containing results in a bunch of csvs, and the associated R commands used.
• Joint pathway analysis - Takes as input a gene list (using official gene names, with optional fold changes) and a metabolite list (KEGG Id, with optional fold changes). Outputs a heatmap of pathway impact (effect size), with associated p-values. Results can be downloaded in a zip file containing results in a bunch of csvs, and the associated R commands used.
• Network explorer - Projects your sample data onto a known network, (e.g. KEGG global metabolic network, HMDB metabolite-disease network, STITCH metabolite-metabolite interaction network...). Works something like KEGG mapper, but has more networks available.
• "MS Peaks to Pathways" - Using input peak list data (optionally with t-scores and p-values), and does pathway enrichment (using GSEA or mummichog) for compound/pathway hits, and network mapping for network exploration on the browser. Annoyingly, GSEA and mummichog can give different p-values for the same dataset (even using their example datasets!).
• Biomarker analysis - Does something with ROC (Recieving operator characteristic) curves; not too sure how this works yet.
• Time-series^ - coming soon
• Power analysis^ - coming soon
• Biomarker meta-analysis - coming soon
• Utilities
  • Chemical compound ID standardization - converts common chemical names to KEGG IDs, HMDB Ids, etc.
  • Batch effect correction - combines multiple datasets (separate .csv or .txt files from different batches) and combines them using ComBat
• Spectral analysis - MetaboAnalysts suggests using xcms or xcms-online for LC-MS, and GC-Autofit for GC-MS.

^ = Data pre-processing steps

These have a fairly standardized, optional, pre-processing procedure preceding the actual analysis:

1. Missing value estimation.
2. Data filtering - filters out variables that are unlikely to be useful. "Highly recommended for untargeted analysis". Non-informative variables are categorized as: a. Variables of very small values, close to baseline or detection limit. b. Variables that are almost constant regardless of experimental groups/conditions. Detected with standard deviation or IQR. c. Variables that show low repeatability - measured using QC samples, with relative standard deviation (RSD), a.k.a. coefficient of variation (CV), where RSD = SD/mean.

• Quite handily, the web app also shows you the R commands that it runs.