ArthOMix
Ask ArthOChat
Omics biomarker discovery platform

From raw omics data to potential biomarkers.

ArthOMix is a Shiny based web application used for transcriptomics, methylomics and integrated platform for identifying potential biomarkers across different omics layers.

Ask ArthOChat
OR GO STRAIGHT TO A MODULE
Enter
Omics coverage
Transcriptomics • Methylomics • Cross-Omics • Multi-omics
Analysis design
Sex-stratified • Sex-specific • Sex-pooled
Accepted data
CSV, RDS file formats.
1
Choose data
2
Choose analysis
3
Check samples
4
Run analysis
5
Explore biomarkers
6
Biomarker modeling
7
Validate
8
Interpret biomarkers
The goal

Choose the analysis.

ArthOMix is designed as a modular application. The user can either upload raw data or normalised data for each of the omics layers. For each module (Transcriptomics, methylomics, Integrated) it has multiple sub-modules. See the upload guide for file format details before you upload the data

Stuck at any step? Ask ArthOChat.

Sex is a design choice here, not an afterthought

Every comparison can be run three ways. Load a dataset and its Dataset tab shows the sample counts per sex, per group, and warns before you commit to a design if a stratum is too small.

Sex-pooled

All samples in one model, adjusted for sex.

Sex-stratified

The same comparison run separately in each sex, so you can compare the two.

Sex-specific

Tests whether the disease effect differs between the sexes - a diagnosis-by-sex interaction model.

Why ArthOMix

One environment for reproducible biomarker discovery.

Multiple analysis

From raw data to potential biomarkers, feature selection, diagnostic modeling, and biological interpretation within one analytical workflow.

Flexible, study-aware analysis

Run sex-stratified, sex-specific, or sex-pooled analysis based on the research design.

Analysis-grounded AI assistant

ArthOChat, grounded in your data

Struggling with a basic concept? Learn while analysing and interpret biological findings with responses grounded in your analysis.

Supports single and multiomics

All in one app, from loading data to preprocessing, statistical analysis, validation, and interpreting results.

Frequently asked

What people ask before analysis.

Do I need to write code to run any analysis?

No. Every step - preprocessing, differential expression or differential methylation analysis, feature selection, model training and evaluation is click and analyse inside the platform's sub-modules. No R or Python is required.

What types of omics data can ArthOMix analyze?

It can analyse Transcriptomics, methylomics, and integrated studies like cross-omics or multiomics.

Can I load my own data?

Yes - use the upload option to upload your own data. This is a public server, so please upload only de-identified data and never identifiable patient information.

Can I perform pooled and sex-specific analyses?

Yes. It can perform sex specific analysis and all the results are downloadable in CSV or PNG file format.

What data does ArthOMix provide?

A platform to analyse omics data. It loads preloaded data as an example, so you can explore the full pipeline before bringing your own data.

Why it is named as ArthOMix?

This shiny based application allows the user to analyse any omics data but since the preloaded data (example) data is confined to Rheumatoid arthritis so the name "Arth", "O" stands for omics and "Mix" because it can perform both single and multiomics analysis based on the data provided.

Ready to explore your data?

Explore the platform's analysis modules to get started.

An example dataset is already loaded, so you can explore right away - or swap in your own data from the Dataset tab.

Developed by the Kovalchuk Lab, University of Lethbridge
For research use only, not for clinical decisions. Please do not upload identifiable patient data to this public server.
ArthOMix v1.0.0

Modules

Each module is one omics layer. Transcriptomics, Methylomics, and Integrated omics. Open one to load a dataset, or run preloadeed data or fetch data from NCBI GEO datasets, add sub-modules and perform analysis

Single-omics modules

Transcriptomics

Available

Differential expression, WGCNA, Mendelian randomisation and colocalisation, feature selection,machine learning modelling and cross-tissue validation, cross-ancestry validation and biomarker card

Methylomics

Available

Differential methylation, WGCNA, Mendelian randomisation and colocalisation, feature selection,machine learning modelling and external validation and biomarker card

Multi-omics modules

Cross-Omics

Available

Evidence integration: combine DEG, DMP/DMR, and eQTL/mQTL-MR results you've already computed to see which genes/CpGs agree across layers.

Multi-Omics

Available

Matched-sample analysis: jointly model raw expression and methylation data from the same subjects with DIABLO/SNF/MOFA.

Assistant

ArthOChat

Available

Ask anything about your dataset, your results so far, or the underlying methodology, grounded in a live PubMed search with citations. Automatically scoped to whichever module/sub-module you have open, with other modules available on request.

Open module
TRANSCRIPTOMICS

Preloaded Datasets

Fetch from NCBI GEO

GEO Series accession (e.g. GSE12345 ). Fetches the expression data and metadata from NCBI. For your own data, use "Upload your own data" instead.

Upload your own data

STEP 1 · Choose your files

Expression matrix - CSV or RDS. Genes in rows, samples in columns; for CSV, the first column is the gene ID.

Sample metadata - CSV or RDS data frame, one row per sample.

STEP 2 · Map the columns
STEP 3 · Confirm
Loads exactly what you provide, as-is - no merging, normalising, or batch correction.

Reserve validation samples (sealed hold-out)

Set aside samples before any analysis runs. Reserved samples are hidden from every step (DE, WGCNA, Candidate Genes, Feature Selection, Model Training). They're only scored later in Diagnostic Model → External Validation, giving the one result that's leakage-safe end-to-end. Note: this turns off the fast precomputed results, since those were built using all samples.



Step 3 · Check samples

See the cohort composition - group and sex counts - for the design just locked in.

Overview and Datasets

Quality control for the currently loaded dataset - preloaded or uploaded.

Preprocessing and Batch Correction

Preprocess and merge one or more datasets based on shared genes and probes, then normalise and batch-correct on the uploaded or pre-loaded data.

Step 4 · Run analysis

Run the chosen analyses on their existing pages.

Differential Expression

Fit a limma or DESeq2 model comparing two levels of a metadata column, optionally stratified by sex (Female/Male).

WGCNA Co-expression Network

Co-expression network analysis: gene/sample filtering, soft-threshold power, module detection, module-trait correlation, hub genes, enrichment, and Cytoscape export.

Mendelian Randomization

Two-sample Mendelian randomisation on preloaded or uploaded data. GWAS summary statistics for any trait are needed.

Colocalization

Bayesian colocalisation testing whether the eQTL and a GWAS's disease-risk association at each candidate locus share a single causal variant.

Sex Interaction Analysis

Diagnosis-by-sex interaction model showing which genes respond to the group difference differently in each sex, plus within-sex disease effects from the same fit.

Immune Deconvolution

Estimate immune cell composition (CIBERSORT/LM22, corroborated with MCP-counter) for pre-loaded expression matrix or uploaded data.

Step 5 · Explore biomarkers

Narrow the results to a candidate feature set and see what it is enriched for.

Candidate Gene Identification

Intersect the disease-associated WGCNA module with the DEG list (sex-stratified when supported), optionally refined by MR/Colocalization support. Visualise and export the Venn diagram and candidate table.

Functional Enrichment

Perform GO, KEGG or Reactome over-representation analysis on a gene list either uploaded or pre-loaded.

Step 6 · Biomarker modeling

Turn the candidates into a consensus panel and train a classifier on it.

Feature Selection

Select features based on LASSO, random forest and SVM-RFE by sex.

Diagnostic Model

Diagnostic model using logistic regression, elastic net, random forest and SVM diagnostic models, by sex.

Step 7 · Validate

Check the panel in an independent cohort, another tissue and another ancestry.

Cross-Tissue Replication

Replicates a blood-derived gene panel in synovial tissue (bundled GSE89408 RNA-seq, or an uploaded cohort), by sex: direction concordance, adjusted p-value, AUC, plus four classifiers refit in the tissue cohort. Replicates the gene-level signal only - not a transfer of the blood-trained diagnostic model.

Cross-Ancestry MR Replication

MR replication of prioritised genes across three GWAS arms (Okada 2014 EUR, Stahl 2010 EUR, BioBank Japan EAS), per sex. Okada includes Stahl, so only BioBank Japan is truly independent. Arms can be replaced by upload. Fixed gene list only - no diagnostic-model or live-dataset use.

Step 8 · Interpret biomarkers

Bring the dataset evidence and every saved result together on the Biomarker Card.

Biomarker Card

Single-gene or gene-panel biomarker lookup: preloaded/uploaded data, live GO/KEGG/Reactome/WikiPathways/Open Targets/HPA/STRING/DGIdb/PubMed queries, downloadable report.

METHYLOMICS

Reference / Example Dataset

Fetch from NCBI GEO

GEO Series accession (e.g. GSE12345 ). Fetches a methylation beta-value matrix and sample metadata from NCBI. For your own data, use "Upload your own data" instead.

Upload your own data

STEP 1 · Choose your files

Methylation matrix - CSV/TSV/RDS. Probes (CpGs) in rows, samples in columns; for CSV/TSV, the first column is the probe ID.

IDAT files : select every _Grn.idat and _Red.idat file at once. Beta values, detection p-values, and bead counts are all derived from these, unnormalized.

Sample sheet / phenotype metadata - CSV/TSV/RDS, one row per sample (optional, but needed to map group/sex/batch below).

STEP 2 · Map the columns
STEP 3 · Confirm
Loads exactly what you provide, as-is - no normalising or batch correction. For that, use this module's own Normalization and Batch Correction sub-modules instead.


Data

Load, quality-control, and prepare the methylation dataset.

Quality Control

Performs probe and sample QC

Normalization

Performs normalization through various methods

Cell-Type Deconvolution

Estimates cell-type proportions from bulk methylation. Works on the loaded dataset, or upload data

Differential Methylation (DMPs)

Performs differentially methylated positions analysis.

Differentially Methylated Regions (DMRs)

Performs differentially methylated region analysis.

Network

Co-methylation structure and the candidate CpGs it points to.

WGCNA (Co-Methylation Network)

Performs co-methylation network analysis

Candidate CpGs (Module-DMR Overlap)

Finds CpGs that overlap significant DMRs and ranks them by WGCNA module enrichment.

Genetics

Causal evidence from mQTL and GWAS summary statistics.

Mendelian Randomization

Two-sample Mendelian randomisation of mQTL instruments against a GWAS outcome.

Colocalisation

Tests whether an mQTL and a GWAS signal at a CpG's region share a causal variant

Biomarker modeling

Turn candidate CpGs into a panel and a diagnostic model.

Sex Interaction Analysis

Diagnosis-by-sex interaction model (fit on M-values) on the loaded data. Shows which CpGs respond to the group difference differently by sex, plus the within-sex disease effects from the same fit.

ML Feature Selection

CpG feature selection using univariate, regularization, tree-based, RFE, and stability selection methods

Diagnostic Classifier

Diagnostic classifiers (logistic regression, elastic net, SVM, random forest, XGBoost, kNN) on single CpGs or a panel, with cross-validated tuning.

Validation

External validation of the Diagnostic Classifier's trained models on an independent cohort

Interpretation

Profile a single candidate biomarker in depth.

Biomarker Card

Shows the potential methylomics biomarker profile.

CROSS-OMICS

Cross-Omics


You're here to combine result tables, not raw data: pull in the DEG table from Transcriptomics and/or the DMP/DMR table from Methylomics that you've already run this session, or upload files in the same format below. Looking to jointly model matched expression + methylation matrices from the same subjects instead? That's the Multi-Omics tab.

1. Data Source

Columns are auto-detected (gene symbol, log2FC/Δβ, P-value, FDR).

DEG format: one row per gene, with a gene symbol/ID and a log2 fold-change column.

DMP/DMR format: one row per CpG or region, with a gene symbol/ID and a Δβ (methylation change) column.

Preview



Data

Join the eQTL-MR/mQTL-MR/DEG/DMP tables into one biomarker-convergence view.

Expression and Methylation

Gene-level integration of Transcriptomics differential expression and Methylomics differential methylation.

Biomarker Convergence

Loads and relabels precomputed eQTL-MR and mQTL-MR evidence at configurable significance thresholds.

Genetics

Bridge the two causal panels with a follow-up cross-omics MR stage.

MR Convergence

Loads and classifies already-computed MR/DEG/DMP/DMR/QTL evidence into 5 named convergence categories (DEG-DMP-QTL, DEG-DMR-QTL, DEG-eQTL, DMP-mQTL, DMR-mQTL). Does not compute MR itself - MR is run externally (preloaded pipeline output, or your own upload of already-run results).

MULTI-OMICS
Ask ArthOChat

Questions about DIABLO, SNF, or how these panels were built? Ask ArthOChat.

Open ArthOChat

Multi-Omics


You're here to load matched, per-subject data: raw expression and methylation measurements from the same samples, for DIABLO/SNF/MOFA to jointly model. Pick a preloaded reference cohort, upload your own matrices, or fetch matched series from GEO below. Looking to compare DEG/DMP result tables you've already computed instead? That's the Cross-Omics tab.

Select Dataset Source

Preloaded

Table

Retrieve from GEO

Enter a GEO SuperSeries accession with matched expression and methylation sub-series; it will be auto-split into both layers.

Many published studies submit expression and methylation as two separate, unlinked GEO series (no SuperSeries relationship) - if the accession above fails with "doesn't look like a SuperSeries", fetch each one directly here instead.

Two independently-submitted series almost never share sample IDs. After fetching, go to the "Sample Matching" tab and use "Patient ID (from metadata)" - if GEO's own per-sample titles end in a shared patient number (a common convention), it is auto-detected as the column "geo_title_patient_num"; otherwise pick whichever metadata column truly identifies the same patient in both series, or supply a mapping file.

Sample Metadata

Imported automatically from each fetched GEO series' own sample metadata.

Data Provenance



Data

Data preparation.

Cohort Harmonization

Modality availability, sample matching, and integration readiness for the active dataset.

Multi-omics Integration (DIABLO & SNF)

DIABLO (supervised) and SNF (unsupervised) integration, with single-omics comparison and sex-stratified analysis.

MOFA

Run a real MOFA2 factor analysis on the Active Multi-Omics Dataset built on the Dataset Workspace tab (matched samples only).

SNF Clustering

Unsupervised patient clustering via Similarity Network Fusion (SNF).

Biomarker modeling

Rank candidate multi-omics biomarkers.

Biomarker Discovery

Supervised DIABLO feature selection across Transcriptomics + Methylomics.

Gene–CpG Mapping

Links candidate gene expression to CpG methylation changes, with direction classification and sex-specific analysis.

Interpretation

Biomarker interpretation.

Pathways

GO / KEGG / Reactome / WikiPathways / Hallmark enrichment on candidate biomarkers or your own upload, with real pathway maps.

Biomarker Card

Integrated gene–CpG biomarker interpretation: expression, methylation, and cross-omics evidence for one candidate at a time.

ArthOChat

Ask about your dataset, results, or the science behind them.