DUET: Dual-Paradigm Adaptive Expert Triage with Single-cell Inductive Prior for Spatial Transcriptomics Prediction

2026-05-15 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, long

Summary

DUET is a novel dual-paradigm framework designed to infer spatially resolved gene expression from histology images, offering a cost-effective alternative to spatial transcriptomics (ST). It addresses limitations of existing methods, which often reduce the task to simple morphology-to-expression mapping and fail to incorporate large-scale single-cell RNA sequencing (scRNA-seq) data. DUET integrates parametric prediction (regression) and memory-based retrieval under cellular inductive priors, adaptively reconciling their outputs. By leveraging millions of scRNA-seq cells, DUET imposes molecular states as biological constraints, mitigating aleatoric vision ambiguity. A lightweight adapter dynamically assigns branch preference across spatial contexts for optimal performance. Extensive experiments on three public datasets (HER2, Breast Cancer, Kidney) across varied gene scales (100, 300, 500 high-variance genes) demonstrate that DUET achieves state-of-the-art performance, with consistent gains from each component.

Key takeaway

For AI Scientists and Machine Learning Engineers developing spatial transcriptomics prediction models, DUET's dual-paradigm approach offers a robust solution to overcome limitations of monolithic methods. You should consider integrating large-scale single-cell references to impose biological constraints and implement an adaptive fusion mechanism to balance expressive flexibility with biological fidelity, potentially improving prediction accuracy and reliability in real-world clinical scenarios.

Key insights

DUET unifies regression and retrieval with single-cell priors for robust spatial gene expression prediction from histology.

Principles

• Morphological similarity does not guarantee molecular consistency.
• Adaptive fusion of paradigms balances flexibility and reliability.
• Large-scale single-cell data provides crucial biological constraints.

Method

DUET constructs cellular inductive priors from scRNA-seq, performs cell-aware gating retrieval, and uses a dynamic soft consistency loss for regression. An adaptive expert triage module then fuses predictions via a lightweight MLP.

In practice

• Utilize scRNA-seq data for cell-type deconvolution.
• Employ a dual regression-retrieval architecture.
• Implement adaptive fusion for pathway preference.

Topics

• DUET Framework
• Spatial Transcriptomics
• Gene Expression Prediction
• Cellular Inductive Prior
• Adaptive Expert Triage

Code references

• Junchao-Zhu/DUET

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Prototype-driven fusion of pathology and spatial transcriptomics for interpretable survival prediction — PathoSpatial fuses WSI and ST data via prototype learning for interpretable, spatially-informed survival prediction.
• Harnessing the power of single-cell large language models with parameter-efficient fine-tuning using scPEFT — scPEFT efficiently adapts single-cell LLMs to new tasks, reducing computational demands and improving performance.
• 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.
• Deep learning-based in silico labeling for analyzing morphological features of MSCs to predict immunomodulatory capacity — A dual AI model non-invasively predicts MSC immunomodulatory capacity from morphology, enhancing cell therapy quality control.
• Defining cancer spatial ecotypes — Spatial EcoTyper uses ML to define nine robust spatial ecotypes in the tumor microenvironment, revealing cell-type-agnostic expression signatures.
• Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement — Cellina disentangles intrinsic cell state from spatial context to predict counterfactual cell expression on tissue graphs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.