MuCoSTX: Multi-Slice Dynamic Anchor-based Contrastive Learning for Spatial Transcriptomics Integration and Alignment

Overview of MuCoSTX

MuCoSTX is a computational framework for multi-slice spatial transcriptomics data integration and alignment, designed to remove systematic batch effects across tissue slices, sample sources, and sequencing platforms while preserving biological manifold structures and local spatial topology. Building on the single-slice spatial representation learning framework of MuCoST, MuCoSTX addresses the challenges of multi-slice ST integration through a macro-micro joint optimization architecture with dynamic anchor contrastive learning.

Quick Start

SEE Notebook Tutorial.

Experimental Results

MuCoSTX evaluated on multiple benchmark datasets (10x Visium DLPFC, Slide-seq V2 MOB, Stereo-seq MDE, etc.) in:

• Batch effect removal: Higher iLISI and lower ASW-batch scores.
• Biological structure preservation: Higher ASW-label, cLISI, and Graph Connectivity scores.
• 3D spatial alignment: Accurate stitching of serial tissue slices for 3D atlas reconstruction.
• Robustness and scalable: Scaleble across different number of samples.

Contact

For questions or issues, please open an issue in the repository or contact [2110610@tongji.edu.cn].