Myofibroblast Stromal State Analysis in Inflammatory Bowel Disease

This repository contains analysis workflows used to quantify and evaluate stromal myofibroblast states in inflammatory bowel disease (IBD).

---

Overview

We identify two opposing stromal programs:

• CXCL14⁺ inflammatory myofibroblasts (IMF)
• CD36⁺ quiescent myofibroblasts (QMF)

These states define a biologically interpretable axis of stromal activation associated with disease severity and outcomes.

---

Methodological Framework

COMPASS (COMPosite Activity Scoring System)

Gene signature activity is quantified using COMPASS, a deterministic, threshold-based scoring framework: https://compass.precsn.com/

Score = (expression − (threshold + 0.5)) / (3 × SD)

• Thresholds derived using StepMiner
• Scores aggregated across genes to generate sample-level activity
• Preserves gene directionality
• No dependence on permutations or gene ontologies

This enables reproducible, interpretable digital biomarkers across datasets.

---

Analyses Included

• Dot plot analysis
  Visualizes signature activity, directionality, and classification strength (AUC)

• Univariate (UV) analysis
  Association testing using t-test or OLS regression

• Multivariate (MV) analysis
  Covariate-adjusted modeling using statsmodels

• ROC–AUC evaluation
  Classification performance assessed using scikit-learn

---

Files

• MF_dot_plots.ipynb Signature visualization
• UV_MV_analysis.ipynb Statistical modeling (UV + MV)
• MF_15gene_test.txt Example dataset

---

Input Format

• Rows: genes
• Columns: samples
• Values: normalized expression (e.g., TPM)

Optional metadata can be included for group labels and clinical variables.

---

Usage

Run analyses directly from notebooks:

1. Open MF_dot_plots.ipynb run all cells
2. Open UV_MV_analysis.ipynb run all cells

---

Reproducibility

• Deterministic scoring (COMPASS)
• Explicit thresholding (StepMiner)
• No stochastic steps
• Fully reproducible from provided data

---

Interpretation

• High IMF score inflammatory, disease-associated state
• High QMF score quiescent, homeostatic state

This framework enables biologically grounded patient stratification and hypothesis generation.

---

Requirements

• Python ≥ 3.8
• pandas
• numpy
• scipy
• matplotlib
• seaborn
• scikit-learn
• statsmodels
• StepMiner (for threshold calculation, if not precomputed)

---

Contact

Saptarshi Sinha
UC San Diego