✦ ISO ResearchMultimodal Research Platform
No analysis has been run yet.

Analyze

Explain

Choose an analysis workflow

Models, single-sample healthy atlas comparison, and multimodal RNA/ATAC/methylation analysis are now grouped here.

Research only: Outputs are exploratory and are not medical advice, diagnosis, or treatment.

Healthy Atlas Compare

Compare one sample against a healthy reference atlas and review deviation, pathway, and top-gene signals.

Use below

Multimodal VAE

Train RNA, ATAC, and methylation together in a shared latent space with disagreement/anomaly outputs.

Open Multimodal ↑

Model Catalog

Review the current model modules and what each one is designed to evaluate.

View Models

Multimodal Shared Latent Model

Aligns RNA, ATAC, and methylation matrices by sample ID, preprocesses each modality, builds a shared latent biological state space, and reports per-modality reconstruction error, disagreement, and research anomaly scores.

Current MVP uses a lightweight shared latent engine; the PyTorch PoE VAE scaffold is included for the next training-service phase.

Health Atlas Comparator

Compares a personal expression sample against a healthy baseline atlas to estimate how far the sample deviates from normal reference biology.

Disease Pattern Similarity

Compares the abnormal portion of a sample against curated biological signatures such as interferon-high viral inflammation, autoimmune-like activation, neutrophil-dominant stress, or metabolic inflammatory shift.

This is a resemblance engine, not a diagnosis engine.

Healthy Atlas Compare

Upload your reference atlas and personal sample

Healthy atlas format: rows = healthy reference samples, columns = genes, values = counts or normalized expression. Personal sample format: either one row with genes as columns, or two columns named gene_id and value.

Healthy atlas

Personal sample