tfbpmodeling

Code style: black style Pytest codecov

A Python package for Transcription Factor Binding and Perturbation (TFBP) modeling that analyzes relationships between transcription factor binding and gene expression perturbations using LASSO.

What is tfbpmodeling?

tfbpmodeling provides a workflow for modeling the relationship between transcription factor binding data and gene expression perturbation data. The package uses bootstrap resampling and regularized regression (LassoCV) to identify significant binding-perturbation relationships while controlling for confounding factors.

Workflow Overview

Stage 0 — Preprocessing

Data validation and preparation: load response and predictor files, apply blacklist, construct the model formula.

Stage 1 — All Data Modeling

Bootstrap LassoCV on the complete dataset using the full interactor model. Significant predictors are extracted at the specified confidence level (--all_data_ci_level). A best all-data model is then fit on those predictors.

Stage 2 — Top-N Modeling

Bootstrap LassoCV on the top-N data subset (genes ranked by perturbed TF binding) using the significant predictors from Stage 1. Surviving predictors are extracted at --topn_ci_level.

Stage 3 — Surviving interactor significance test

Tests each surviving interactor term against its corresponding main effect to determine whether the interaction provides genuine predictive value beyond the main effect alone. Uses linear regression by default; pass --stage3_lasso to use LassoCV instead. Pass --stage3_lasso_topn to run the test on the top-N data subset rather than all data.

Stage 3 - LassoCV Bootstrap (--stage3_lassocv_bootstrap, optional)

Independently of the significance test, refits the surviving interactors together with their independent main effects on all data using the same bootstrap LassoCV protocol as Stage 1.

Citation

If you use tfbpmodeling in your research, please cite:

@software{tfbpmodeling,
  title = {tfbpmodeling: Transcription Factor Binding and Perturbation Modeling},
  author = {Mateusiak, Chase and Erdenebaatar, Zolboo and Eric Jia and Mueller, Ben and Liu, Chenxing and Brent, Michael},
  url = {https://github.com/BrentLab/tfbpmodeling},
  year = {2025}
}

Support