MIRACLE Documentation

MIRACLE is a continual integration framework for single-cell multimodal data.

Core Capabilities

• Continual multimodal integration across sequential tasks and data releases.

• Replay memory construction for preserving previously learned biological structure during incremental updates.

• Feature alignment when new measurements add genes, proteins, peaks, or other modality-specific variables.

• Checkpoint loading, compatible parameter transfer, and latent extraction for downstream Scanpy/MuData workflows.

Contents:

• Installation
  • Create an Environment
  • Verify the API
  • GPU Notes
• Quick Start
  • Imports
  • Data Contract
  • Initial Training
  • Build Replay Memory
  • Continual Integration
  • Important Constraints
  • Inference from a Checkpoint
• Tutorials
  • Tutorial 1: Continual Integration of Multimodal Mosaic Single-Cell Data
  • Tutorial 2: Label Transfer from a Reference Atlas to a Query Dataset
• API Reference
  • MIRACLE API
• Release Notes
  • 0.1.0