Analyze / 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.

Interpretability Layer

Turns raw output scores into plain-language biological explanations so the user can understand whether the signal looks more like immune activation, pathway dysregulation, or blood-cell composition shift.