LATTICE: Graph Self-Supervised Learning for Multimodal Spatial Omics Integration

2026-07-15 · Source: Machine Learning · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

LATTICE (Latent Alignment of Tissue-level and Transcriptomic Information for Cross-modal Embedding) is a novel graph-based self-supervised framework designed for multimodal spatial omics integration. It learns spot-level representations by harmonizing features from five aligned modalities per Visium spot: Visium RNA, scMultiome RNA, scMultiome ATAC, spatial ATAC, and spatial CUT&Tag. The framework constructs a spatial neighborhood graph and employs a TransformerConv encoder, trained with masked reconstruction, cross-modal alignment, and spatial smoothness objectives. Evaluated on a private 11-sample melanoma cohort comprising 54,912 spots, LATTICE demonstrated stable optimization and reproducible embeddings. Integrating scMultiome RNA with Visium RNA alone significantly improved concordance with Space Ranger clusters (adjusted Rand index +0.157, normalized mutual information +0.143, spatial contiguity +0.174). Further modalities enhanced spatial contiguity and multimodal utility score, indicating the embeddings captured regulatory and chromatin structures beyond just transcriptomic similarity.

Key takeaway

For research scientists integrating multimodal spatial omics data, LATTICE offers a robust graph-based self-supervised framework to harmonize diverse assays. You should consider adopting this approach to generate more comprehensive spot-level representations, especially when combining Visium RNA with scMultiome RNA. This method can reveal chromatin and regulatory structures beyond transcriptomic similarity, providing richer biological context for your analyses. Further benchmarking is recommended.

Key insights

LATTICE integrates diverse spatial omics data using graph self-supervision for robust spot-level representations.

Principles

• Multimodal integration enhances spatial omics analysis.
• Graph-based self-supervision can harmonize diverse biological data.
• Embeddings can capture regulatory structure beyond transcriptomics.

Method

LATTICE constructs a spatial neighborhood graph, then trains a TransformerConv encoder using masked reconstruction, cross-modal alignment, and spatial smoothness objectives to learn spot-level representations.

In practice

• Combine Visium RNA with scMultiome RNA for improved clustering.
• Utilize graph-based self-supervision for complex omics data.
• Explore chromatin and regulatory structures via multimodal embeddings.

Topics

• Spatial Omics Integration
• Graph Self-Supervised Learning
• Multimodal Omics
• Visium RNA
• scMultiome ATAC
• TransformerConv

Best for: AI Scientist, Research Scientist

Related on AIssential

• scLLM-DSC: LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering for Single-Cell RNA Sequencing — scLLM-DSC integrates LLM-derived biological semantics with transcriptomic features for superior scRNA-seq clustering.
• Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement — Cellina disentangles intrinsic cell state from spatial context to predict counterfactual cell expression on tissue graphs.
• DUET: Dual-Paradigm Adaptive Expert Triage with Single-cell Inductive Prior for Spatial Transcriptomics Prediction — DUET unifies regression and retrieval with single-cell priors for robust spatial gene expression prediction from histology.
• Systematically decoding pathological morphologies and molecular profiles with unified multimodal embedding — Multi-Embed unifies morphological and molecular data for superior disease understanding and prediction.
• OMIDIENT: Multiomics Integration for Cancer by Dirichlet Auto-Encoder Networks — OMIDIENT integrates multi-omics cancer data using Dirichlet auto-encoder networks for superior unsupervised analysis and interpretability.
• Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors — Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.