Help & User Guide

EV Barcode predicts the cellular source of extracellular vesicles (EVs) based on their cargo composition, enabling comparative analysis and visualization of EV-associated gene expression across biological conditions and cell types. This guide outlines input formats, data submission workflows, normalization steps, and output features.

EV Barcode is intended for research use. It is not a clinically validated diagnostic system. Predictions depend on reference-atlas coverage and may be less informative for rare, absent, underrepresented or substantially altered disease cell states.

EV Barcode analysis workflow — EV Barcode workflow.

Accepted Input Formats

EV Barcode supports multiple file formats to ensure compatibility with common bioinformatics workflows:

Plain text files (.txt)
Microsoft Excel files (.xls, .xlsx)
Comma-separated values (.csv)

The platform accepts the following gene identifier types:

Gene IDs
Ensembl IDs
Gene Symbols

All identifiers can be automatically mapped using the built-in gene mapping function to ensure consistent annotation. The mapping preview reports total submitted identifiers, mapped and unmapped identifiers, mapped percentage, duplicate identifiers collapsed, and the identifiers eligible for analysis.

Data Submission Methods

Real-Time Query (Query Tab)

The Query tab is optimized for smaller datasets and starts analysis immediately through the EV Barcode processing queue. This mode is recommended for exploratory use and rapid data inspection.

Designed for small gene lists
Fast queued analysis for suitable inputs
Email notification is optional if processing takes longer than expected

Data Submission (Submit Data Tab)

The Submit Data tab supports large datasets and asynchronous processing. It is suitable for high-throughput analyses or delayed workflows.

Optimized for large gene lists
Asynchronous processing
Email notification upon completion when an email address is provided

Each upload receives a UUID that can be used to retrieve results later from the View Submission page.

Analysis and Interpretation

Transcriptomic and proteomic EV cargo files are accepted and analysed as separate submissions. EV Barcode does not numerically merge RNA and protein values into a single multi-omics probability.

The workflow is: select species and molecular modality, upload EV cargo, validate the file and map identifiers, collapse duplicate mapped identifiers, exclude unmapped features, review mapped and unmapped inputs, run the implemented analytical methods, then display predicted cell types with method-specific statistics and supporting gene-level evidence.

The Gene Specificity Index supports cell-type-specific evidence. The Ubiquity Index uses expression threshold t = 0.15; for each gene, UI is the proportion of cell-type reference signatures where normalized aggregated expression exceeds that threshold. UI near 0 indicates restricted expression, while UI near 1 indicates broad expression. Genes with UI > 0.85 are excluded from cell-type-discriminatory specificity scoring, and retained genes are down-weighted by (1 - UI).

For each predicted cell type, the hypergeometric test reports N, Kc, n, kc and a one-sided upper-tail p-value. N is the species-specific eligible reference universe; Kc is the number of genes in that universe annotated as markers for that cell type; n is the number of unique mapped input genes eligible for analysis; kc is the number of mapped input genes identified as markers of that cell type. A smaller p-value means stronger enrichment evidence, not the probability that a predicted cell type is the true source.

Marker identification, weighted cell-type gene score, GSEA-like enrichment, NNLS, NMF, cargo similarity and supporting gene-level evidence have different meanings and scales. They are not added together as a calibrated confidence probability. A smaller p-value, a higher weighted score and consistent support across analytical methods provide stronger statistical and biological evidence for the predicted cellular origin.

EV Score: X/100 reports how many unique mapped submitted features are represented among the EV-associated top 100 genes supplied with this project. It indicates EV-cargo representation and does not represent cellular-origin probability or prediction confidence.

Group Comparison

EV Barcode enables comparative analysis across experimental groups to highlight biologically meaningful differences.

Comparison of two or more groups is supported
A maximum of four groups may be compared simultaneously
This limitation ensures clarity in visualization and interpretation

Expression Values and Replicates

Users may submit expression data in one of the following formats:

Fold values
Normalized expression values (e.g., NSPC)
Raw or processed count data

For experiments containing multiple replicates:

Normalized replicate values may be submitted directly
Alternatively, average values per group may be provided

Upon submission, EV Barcode automatically normalizes data per gene and per condition, depending on the dataset structure and number of comparison groups.

Example Input Table Formats

Below are example table formats illustrating different ways gene-level data can be submitted for EV Barcode analysis.

Gene Symbol
Gene1
Gene2
Gene3
Gene4
Gene5
Gene6

Table 1: Gene list input format without quantitative values.

Gene Symbol	Quantitative Value
Gene1	12
Gene2	14
Gene3	12
Gene4	18
Gene5	13
Gene6	18

Table 2: Single-condition input format using quantitative values (e.g., fold, NSPC, normalized expression, or count-based measures).

Gene Symbol	Control	Group 1	Group 2
Gene1	11	20	20
Gene2	20	14	23
Gene3	11	5	23
Gene4	18	14	6
Gene5	19	9	13
Gene6	16	23	6

Table 3: Multi-group quantitative input format for comparative analysis.

Output and Visualization

EV Barcode generates interactive, high-resolution visualizations combined with comprehensive data tables.

Interactive figures with selectable and deselectable cell types
Dynamic updates to facilitate exploration and interpretation
Publication-ready visual outputs

Users can export both figures and data tables and share results securely with collaborators using a unique access link.

Case Studies

The Query page includes three case studies with downloadable example inputs: immune-cell human EV cargo, mouse plasma EV cargo, and a mouse plasma EV group comparison. The second case study uses mouse plasma EVs isolated in our laboratory and compares predicted cellular sources with experimental findings reported by Vanderboom et al. This is a cross-species, experimentally anchored in silico concordance assessment, not direct lineage validation.

Data Retention and Automated Maintenance

To ensure efficient storage management while maintaining accessibility:

Results from real-time queries are retained for 5 days
Results from data submissions are retained for 7 days

Each submission is assigned a Universally Unique Identifier (UUID), enabling secure access, traceability, and collaboration.

Due to the storage-intensive nature of EV Barcode, an automated maintenance routine runs daily according to the configured Celery beat schedule. This process removes expired results, optimizes storage usage, and preserves overall system performance.