Usage

Web Application

Basic Workflow

1. Create a project – Click “New project” in the Projects page
2. Upload datasets – In the Data tab, upload your training files (Xtrain.tsv, Ytrain.tsv) and optionally test files (Xtest.tsv, Ytest.tsv). You can also pick from the dataset library or load demo datasets.
3. Explore data – Use the Data Explorer to inspect feature statistics, prevalence distributions, volcano plots, and barcode visualizations. Apply feature filtering (Wilcoxon, t-test, Bayesian Fisher).
4. Configure parameters – In the Parameters tab, choose your algorithm (GA, Beam, MCMC), model language (binary, ternary, ratio), and adjust settings like population size, max epochs, k range, and cross-validation folds. Use templates for preset configurations.
5. Launch analysis – Click “Launch Analysis”. Monitor progress in real time via the console panel with live sparkline chart.
6. Explore results – Once completed, browse:
   • Summary: best AUC, k, timing, generation tracking charts
   • Population: feature heatmap, violin plots, prevalence analysis
   • Jury: ensemble voting, confusion matrices, vote matrix, sample predictions
   • Comparative: compare multiple jobs side-by-side
   • Co-presence: feature co-occurrence analysis
   • Ecosystem: co-abundance network with taxonomic coloring and module detection
   • Stability: model stability indices (Kuncheva, Tanimoto, CW_rel), feature × sparsity heatmap, model clustering dendrogram

Batch Mode

In the Parameters tab, enable “Batch Mode” to sweep across multiple configurations:

• Seeds (e.g., 42, 123, 456)
• Algorithms (GA, Beam, MCMC)
• Languages (bin, ter, ratio)
• Data types (raw, prev)
• Population sizes, max epochs, k_max values

The system generates all combinations and launches them as separate jobs (up to 50 per batch).

External Validation

After training, validate your model on an independent cohort:

1. In the Results tab, select a completed job
2. Click “Validate on New Data”
3. Upload the validation X matrix (and optionally Y labels)
4. View AUC, accuracy, confusion matrix, and per-sample predictions

Prediction API

Deploy trained models as REST endpoints:

curl -X POST http://localhost:8001/api/v1/projects/{id}/jobs/{job_id}/predict \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"features": {"species_A": 0.12, "species_B": 0.05}}'

Exporting Results

From the Results tab, click “Export” to access:

• PDF biomarker report (publication-ready)
• HTML report (self-contained)
• CSV files (best model, population, generations, jury predictions)
• Python notebook (.ipynb)
• R notebook (.Rmd)
• Full JSON

Python (gpredomicspy)

Basic Analysis

import gpredomicspy

# Load parameters from YAML
param = gpredomicspy.Param()
param.load("params.yaml")

# Run the evolutionary search
experiment = gpredomicspy.fit(param)

# Display results with jury voting
experiment.display_results()

Accessing Results

# Best individual from the population
best = experiment.best_population().best()
print(best.get_metrics())     # AUC, accuracy, sensitivity, specificity
print(best.get_features())    # Feature names and coefficients
print(best.get_k())           # Number of features

# Generation tracking
tracking = experiment.generation_tracking()

# Jury results
jury = experiment.jury_results()

Parameter Configuration

param = gpredomicspy.Param()
param.set_algorithm("ga")           # ga, beam, mcmc
param.set_language("ter")           # bin, ter, ratio
param.set_data_type("raw")          # raw, prev
param.set_max_epochs(200)
param.set_population_size(100)
param.set_k_range(3, 15)
param.set_n_folds_outer(5)
param.set_seed(42)
param.set_compute_importance(True)
param.set_voting(True)

R Package

library(predomics)

# Load data
data <- loadData("Xtrain.tsv", "Ytrain.tsv")

# Run analysis
result <- mainFunction(data,
  language = "ter",
  algorithm = "ga",
  populationSize = 100,
  maxEpochs = 200
)

# View results
print(result)
plotResults(result)