HistoLoop: Iterative Contextual Inference with a Looped Transformer for Histology-Based Spatial Gene Expression Prediction

Ziang Liu; Jie Liu; Shuai Li; Yigui Feng; Qinglin Wang

cs.CVOct 20, 2026

HistoLoop: Iterative Contextual Inference with a Looped Transformer for Histology-Based Spatial Gene Expression Prediction

Authors: Ziang Liu, Jie Liu, Shuai Li, Yigui Feng, Qinglin Wang

Organizations: Laboratory of Digitizing Software for Frontier Equipment, National University of Defense Technology · National Key Laboratory of Parallel and Distributed Computing, National University of Defense Technology · College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China

Abstract

Predicting spatial gene expression from H&E images offers a scalable complement to experimental spatial transcriptomics. However, despite progress in spatial context modeling, how to construct subsequent context from spot states updated by earlier contextual interactions remains underexplored. We introduce HistoLoop, a looped Transformer that formulates prediction as iterative contextual inference. At each refinement step, HistoLoop aggregates the current spot states into context tokens at the corresponding resolution, processes them with a shared spatial Transformer, and retrieves spot-specific contextual information through cross-attention. A memory-mediated update integrates this information into persistent spot states, which are then used to construct context at the next, finer resolution. Under the HEST-Bench HVG50 evaluation protocol, HistoLoop achieves the highest macro-averaged gene-wise Pearson correlation among the compared methods across six cancer-specific tasks spanning five organs in HEST-1K. Ablation studies support the importance of recurrent state refinement and reconstructing context from updated spot states at progressively finer resolutions. Expression-derived clustering further suggests that HistoLoop better captures spatially organized multigene patterns.

Figures & tables

Figure 1: Overview of HistoLoop. (a) HistoLoop fuses pretrained histological features and 2D spatial coordinates, applies a full-resolution pre-spatial Transformer, initializes persistent spot states, performs

T

recurrent refinement steps following a coarse-to-fine context-token ratio schedule, and predicts gene expression after a full-resolution post-spatial Transformer. (b) At refinement step

t

, the current spot states

\mathbf{H}^{(t)}

are aggregated into context tokens

\mathbf{C}^{(t)}

, which are processed by a shared spatial Transformer to obtain contextualized representations

\mathbf{Z}^{(t)}

. Spot-conditioned cross-attention then retrieves contextual information from

\mathbf{Z}^{(t)}

for individual spots, yielding contextual updates

\Delta\mathbf{H}^{(t)}

. A MeSH-based write–read transition produces the next states

\mathbf{H}^{(t+1)}

, which are used for subsequent context construction.

Cancer Type	HiST	HistoPrism	MoLF	STFlow	HistoLoop
COAD	$0.329_{\scriptstyle 0.011}$	$0.267_{\scriptstyle 0.025}$	$0.273_{\scriptstyle 0.014}$	$0.343_{\scriptstyle 0.019}$	$\mathbf{0.372}_{\scriptstyle 0.008}$
IDC	$0.411_{\scriptstyle 0.004}$	$0.356_{\scriptstyle 0.002}$	$0.609_{\scriptstyle 0.007}$	$0.518_{\scriptstyle 0.031}$	$\mathbf{0.653}_{\scriptstyle 0.005}$
LUNG	$0.306_{\scriptstyle 0.029}$	$0.311_{\scriptstyle 0.021}$	$\mathbf{0.513}_{\scriptstyle 0.019}$	$0.414_{\scriptstyle 0.024}$	$0.470_{\scriptstyle 0.007}$
PAAD	$0.322_{\scriptstyle 0.026}$	$0.252_{\scriptstyle 0.020}$	$0.354_{\scriptstyle 0.023}$	$0.405_{\scriptstyle 0.047}$	$\mathbf{0.479}_{\scriptstyle 0.011}$
READ	$0.263_{\scriptstyle 0.017}$	$0.286_{\scriptstyle 0.006}$	$0.239_{\scriptstyle 0.016}$	$0.266_{\scriptstyle 0.007}$	$\mathbf{0.287}_{\scriptstyle 0.006}$
SKCM	$0.447_{\scriptstyle 0.010}$	$0.386_{\scriptstyle 0.049}$	$\mathbf{0.622}_{\scriptstyle 0.009}$	$0.481_{\scriptstyle 0.015}$	$0.621_{\scriptstyle 0.004}$

Table 1: Comparison of HVG50 PCC on the six cancer-specific tasks. Each cancer-specific entry reports the mean over five random seeds, with the standard deviation shown as a lower-right subscript. Macro Avg. is the unweighted mean of the six cancer-specific mean PCCs. The best mean in each row is shown in bold.

Variant	Params	COAD	IDC	LUNG	PAAD	READ	SKCM	Macro Avg.
Full HistoLoop	674.8	$0.372_{\scriptstyle 0.008}$	$0.653_{\scriptstyle 0.005}$	$0.470_{\scriptstyle 0.007}$	$0.479_{\scriptstyle 0.011}$	$0.287_{\scriptstyle 0.006}$	$0.621_{\scriptstyle 0.004}$	0.480
No loop	477.6	$0.325_{\scriptstyle 0.005}$	$0.508_{\scriptstyle 0.008}$	$0.384_{\scriptstyle 0.012}$	$0.379_{\scriptstyle 0.014}$	$0.202_{\scriptstyle 0.009}$	$0.487_{\scriptstyle 0.004}$	0.381
No loop, matched	679.0	$0.321_{\scriptstyle 0.006}$	$0.519_{\scriptstyle 0.004}$	$0.388_{\scriptstyle 0.016}$	$0.397_{\scriptstyle 0.009}$	$0.216_{\scriptstyle 0.012}$	$0.473_{\scriptstyle 0.005}$	0.386
Single scale	674.8	$0.351_{\scriptstyle 0.006}$	$0.617_{\scriptstyle 0.007}$	$0.448_{\scriptstyle 0.013}$	$0.453_{\scriptstyle 0.021}$	$0.272_{\scriptstyle 0.008}$	$0.603_{\scriptstyle 0.003}$	0.457
Reverse order	674.8	$0.346_{\scriptstyle 0.011}$	$0.622_{\scriptstyle 0.005}$	$0.446_{\scriptstyle 0.008}$	$0.485_{\scriptstyle 0.016}$	$0.275_{\scriptstyle 0.007}$	$0.603_{\scriptstyle 0.006}$	0.463
Parallel shared	677.9	$0.338_{\scriptstyle 0.006}$	$0.632_{\scriptstyle 0.004}$	$0.443_{\scriptstyle 0.013}$	$0.491_{\scriptstyle 0.012}$	$0.272_{\scriptstyle 0.009}$	$0.591_{\scriptstyle 0.007}$	0.461

Table 2: Ablation study of HistoLoop on the six cancer-specific tasks. Each cancer-specific entry reports the mean HVG50 PCC over five random seeds, with the standard deviation shown as a lower-right subscript. Parameter counts are reported in millions. Macro Avg. is the unweighted mean of the six cancer-specific mean PCCs. The best mean in each cancer column is shown in bold.

Figure 2: Spatial expression recovery on representative held-out slides.

Model	ARI $\uparrow$	AMI $\uparrow$	NMI $\uparrow$	Homo. $\uparrow$
HiST	0.319	0.479	0.482	0.522
HistoPrism	0.224	0.377	0.381	0.429
MoLF	0.276	0.433	0.436	0.468
STFlow	0.211	0.360	0.364	0.406
HistoLoop	0.329	0.526	0.530	0.611

Table 3: Overall clustering consistency between ground-truth and predicted expression profiles.

Figure 3: Expression-derived clustering on two held-out slides.

Spatial Gene Expression Prediction

Spatial gene expression prediction infers tissue-resolved molecular activity from histology and spatial context.

See all 39 papers

MIST: Multi-Scale Spatial Context for Gene Expression Prediction from Histology
Oct 13, 2026Koushik Howlader, Tirtho Roy, Md Tauhidul Islam +1Spatial Context Modeling Spatial Gene Expression Prediction
Iowa State University, Iowa, USA · Stanford University, CA, USA
HistoGPA: A Context-Conditioned Gene-Prior Attention Framework for Histology-Based Spatial Gene Expression Prediction
Jul 27, 2026Ziang Liu, Xinhai Chen, Yigui Feng +3Computational Pathology Spatial Transcriptomics
Laboratory of Digitizing Software for Frontier Equipment, National University of Defense Technology · National Key Laboratory of Parallel and Distributed Computing, National University of Defense Technology · College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China
Biology-Guided Context-Aware Residual Propagation for Spatial Gene Expression Prediction from Histology
Oct 26, 2026Vu Van An, Long D. Nguyen, Ho Viet Duc Luong +1Spatial Gene Expression Prediction Gene Expression Prediction from Histopathology Images
Hanoi University of Science and Technology Hanoi, 10000, Vietnam · Victoria University of Wellington Wellington, 6012, New Zealand

Oct 13, 2026J/K move · Enter open · S save

MIST: Multi-Scale Spatial Context for Gene Expression Prediction from Histology

Koushik Howlader, Tirtho Roy, Md Tauhidul Islam, Wei Le

Iowa State University, Iowa, USA · Stanford University, CA, USA

Spatial Context Modeling Spatial Gene Expression Prediction Gene Expression Prediction from Histopathology Images Spatial Transcriptomics Multi-Scale Attention Spatial Reasoning

Open paper AskPDF arXiv

Predicting spatial gene expression from standard H&E images could extend molecular analysis to tissues without spatial transcriptomic measurements. A standard Transformer applies self-attention over spots but ignores where each spot sits in the tissue, whereas a spot's molecular state depends on tissue context at multiple spatial scales. Building on this insight, we introduce MIST, a Multi-context Spatial Transformer that augments the self-attention backbone with two context pathways applied in every layer: a coordinate-invariant local pathway that attends over spatial k-nearest-neighbor spots, and a slide pathway that broadcasts a whole-slide summary to every spot; the model predicts all spots in a single forward pass. Because coordinates enter only through pairwise distances, the local pathway is invariant to translations, rotations, and reflections when image features are fixed. In intra-cohort evaluation across six human organ cohorts from HEST-1k, MIST improves mean Pearson correlation over the strongest evaluated baseline by 9.7%, 11.6%, and 11.4% for 10-, 50-, and 100-gene panels, respectively, with the lowest average MSE and MAE. MIST further improves leave-one-organ-out transfer by 6.0% over the next strongest baseline, attains the best expression-derived domain recovery (highest mean ARI and NMI at every clustering resolution, and higher NMI than the closest baseline on 84% of held-out slides), and reproduces spatial expression patterns of selected cancer-marker genes. Together, these results show that adding local and slide context to a vanilla Transformer backbone is an effective approach to spatial gene expression prediction from histology.

Abstract

Spatial Gene Expression Prediction

Explore similar work

MIST: Multi-Scale Spatial Context for Gene Expression Prediction from Histology

HistoGPA: A Context-Conditioned Gene-Prior Attention Framework for Histology-Based Spatial Gene Expression Prediction

Biology-Guided Context-Aware Residual Propagation for Spatial Gene Expression Prediction from Histology