Related Experiment Video
Updated: Aug 5, 2026
06:25
Long-Term Imaging of Identified Neural Populations using Microprisms in Freely Moving and Head-Fixed Animals
Published on: January 19, 2024
Continual integration of single-cell multimodal data with MIRACLE
Jiahao Zhou1,2, Jing Wang1, Shuofeng Hu1
1Center for Computational Biology, Beijing Institute of Basic Medical Sciences, Beijing, China.
Nature Computational Science
|July 31, 2026
Summary
We developed MIRACLE, a continual learning framework for integrating multimodal single-cell data. This scalable approach efficiently updates biomedical atlases with new data, improving research adaptability and knowledge exploration.
Area of Science:
- Computational Biology
- Genomics
- Bioinformatics
Background:
- Single-cell sequencing advances cellular heterogeneity studies and multi-omics atlas construction.
- Current atlas updates necessitate full dataset reintegration, posing scalability issues for biomedical research.
Purpose of the Study:
- Introduce MIRACLE (multimodal integration with continual learning), an online learning framework for scalable multimodal data integration.
- Address the limitations of conventional atlas updates by enabling continuous, efficient integration of diverse datasets.
Main Methods:
- Employ dynamic architecture adaptation and data rehearsal for continual learning.
- Develop an online learning framework for integrating multimodal single-cell data.
Main Results:
- MIRACLE achieves accurate online integration with significantly improved efficiency.
- The framework refines and expands atlases with new cross-modal, cross-tissue, and cross-disease data.
- Analysis of respiratory infections reveals shared and pathogen-specific immune mechanisms in COVID-19, influenza A, and tuberculosis.
Conclusions:
- MIRACLE offers an efficient and collaborative solution for the continual integration, sharing, and exploration of biological knowledge.
- The framework enhances the timeliness and adaptability of biomedical research through scalable atlas updates.
