Framework for Explainable Predictive Oncology and Target Prioritization • powerup

Framework for Explainable Predictive Oncology and Target Prioritization

POWERUP is a framework for building explainable predictive models of therapeutic vulnerabilities from molecular profiles.

Overview

The POWERUP R package provides a standardized and accessible workflow for training machine learning models, evaluating model performance, and generating predictions in new samples. It includes tools for interpreting the molecular features contributing to predictions and integrating new experimental measurements using Bayesian updating. POWERUP also generates analysis-ready summaries and visualizations to support target prioritization and downstream experimental validation.

Installation

Install POWERUP directly from GitHub:

install.packages("remotes")
remotes::install_github("hts-lab/powerup")

See Installing POWERUP for requirements and further setup notes.

Main workflow

Step	Purpose	Main functions
Prepare	Assemble reference and prediction data	`prepare_powerup_data()`
Train	Train and evaluate predictive models	`fit_powerup_models()` `summarize_models()`
Predict	Generate sample-level predictions	`add_powerup_predictions()` `summarize_predictions()`
Explain	Examine important features contributing to sample-level predictions and potential diversity in explanations across samples	Local features `summarize_contributions()` `plot_contributions_to_sample()` Explanation diversity `calculate_explanation_paths()` `plot_explanation_paths()`
Observe	Process experimental observations	`prepare_powerup_observations()`
Update	Combine predictions with observations	`calculate_powerup_posteriors()` `plot_posterior()`

Documentation

Start with Get Started with POWERUP. The focused guides cover:

Individual functions are documented in the reference index. Data provenance and acknowledgement information for the bundled examples are documented in Example data and provenance.

Citation

If you use POWERUP in your research, please cite the associated manuscript:

Single tumor transcriptional dependency inference prioritizes rare cancer targets

Mushriq Al-Jazrawe, Merve Dede, Neha Nanda, Colin Trepicchio, Grace Johnson, Kathryn Cebula, Elisabeth Abeyta, Nicole Ostrovsky, Kailee A. Rutherford, Hong Phuc Vo, James Neiswender, Lisa Brenan, Mike Burger, Barbara Van Hare, Candace Haddox, Alice Soragni, Francisca Vazquez, and Jesse S. Boehm.

Publication details: TBD.

The citation is also available from R:

citation("powerup")

For a published analysis, record the POWERUP version:

packageVersion("powerup")

Source and issues

Source code: https://github.com/hts-lab/powerup
Report issues: https://github.com/hts-lab/powerup/issues

POWERUP Portal

The HTS POWERUP Portal provides an interactive interface for running analyses without directly using the R package. The portal additionally orchestrates large analyses (e.g. training predictive models for thousands of perturbations, or inferring vulnerabilities for hundreds of new samples) that would otherwise take too long or require too much memory to complete using this R package alone.

To request access to the portal, contact hts-lab@mit.edu.