STRATA treats regulon activity scores as continuous scalar fields on tissue manifolds, enabling gradient computation, divergence analysis, and field coupling quantification directly on spatial transcriptomics coordinates.
Standard spatial transcriptomics tools focus on cell-type deconvolution and ligand??receptor inference. STRATA provides a complementary, field-theoretic perspective: it computes smooth regulon activity fields from discrete spot-level data, then applies differential operators (gradient, divergence, Laplacian) to quantify spatial architecture.
Key capabilities:
- Compute regulon activity fields on any spatial platform (Visium, MERFISH, Slide-seq, Xenium)
- Data-driven tissue axis definition (e.g., dermal??epidermal junction) without anatomical annotation
- 1D field profiles with bootstrap confidence intervals
- Sample-level phase diagrams for disease classification
- Field coupling analysis between regulatory programs
# Clone
git clone https://github.com/jengweitjiu/STRATA.git
cd STRATA
# Dependencies
pip install numpy scipy pandas scanpy matplotlib decoupler-py omnipathRequirements: Python ??3.9
import pandas as pd
import numpy as np
from scipy.ndimage import gaussian_filter1d
# Load STRATA fields (precomputed regulon activity)
tf_activity = pd.read_csv("tf_activity_spatial.csv", index_col=0)
metadata = pd.read_csv("metadata.csv", index_col=0)
coords = pd.read_csv("spatial_coords.csv", index_col=0)
# Define DEJ axis from regulon fields
epi_tfs = ["KLF4", "SOX9", "TP53", "MYC"]
der_tfs = ["SMAD3", "FOXP3", "SPI1", "IRF1"]
epi_score = tf_activity[epi_tfs].mean(axis=1)
der_score = tf_activity[der_tfs].mean(axis=1)
# Weighted centroids ??axis direction ??project spots
epi_mask = epi_score > epi_score.median()
der_mask = der_score > der_score.median()
epi_centroid = coords.loc[epi_mask, ["imagerow", "imagecol"]].mean()
der_centroid = coords.loc[der_mask, ["imagerow", "imagecol"]].mean()
axis_vec = (epi_centroid - der_centroid).values
axis_vec /= np.linalg.norm(axis_vec)
# Project and normalize to [0, 1]
positions = coords[["imagerow", "imagecol"]].values @ axis_vec
dej_position = (positions - positions.min()) / (positions.max() - positions.min())We validated STRATA on 30 skin biopsies (16,424 Visium spots) from healthy, non-lesional, and lesional psoriasis tissue.
| Finding | Value | Significance |
|---|---|---|
| Inflammatory??Regulatory coupling | ? = 0.94, R² = 0.95 | P = 1.7 ? 10??¹⁹ |
| Coupling slope | 1.06 | Near-perfect 1:1 scaling |
| Regulatory residual vs PASI | ? = 0.47 | P = 0.023 |
| Raw inflammatory vs PASI | ? = 0.04 | P = 0.86 (n.s.) |
| KLF4 bins significant (H vs L) | 34/50 | Bonferroni-corrected |
| FOXP3 bins significant (H vs L) | 0/50 | No spatial difference |
| MYC bins significant | 45/50 | Bonferroni-corrected |
| STAT3 bins significant | 46/50 | Bonferroni-corrected |
Key insight: The regulatory arm scales 1:1 with inflammation ("coupled but overwhelmed"). The imbalance between regulatory and inflammatory fields ??not absolute inflammation ??predicts clinical severity (PASI). This finding is invisible to standard differential expression.
(a) DEJ axis on tissue coordinates. (b) 1D regulon field profiles with 95% bootstrap CIs. (c) Inflammatory??regulatory coupling (R² = 0.95) and disease phase space. (d) Regulatory residual predicts PASI (? = 0.47).
STRATA/
?????? README.md
?????? LICENSE
?????? figures/
?? ?????? Fig_COMPOSITE_v3_reordered.png
?????? (analysis scripts in Zenodo archive)
Full data, scripts, and figures: Zenodo archive
| Dataset | Source | Use |
|---|---|---|
| GSE173706 | Reynolds et al., Science 2021 | scRNA-seq (83,352 cells) |
| GSE202011 | Castillo et al., Sci. Immunol. 2023 | Spatial transcriptomics (16,424 spots) |
- SICAI ??Stromal-Immune Coupled Attractor Index for psoriasis (Zenodo)
@article{tjiu2026strata,
title={STRATA: Spatial Transcription-factor Regulatory Architecture for tissue field analysis},
author={Tjiu, Jeng-Wei},
journal={bioRxiv},
year={2026},
doi={10.64898/2026.02.24.707661}
}MIT License. See LICENSE.