⚡ Fun fact: This landing page will be a work in progress for a while.
Standard LASSO is statistically elegant but biologically blind — in collinear omics data, it will happily select a gene simply because it lives next door to the real driver. BioPrimeLASSO seeks to circumvent this false positive by incorporating exterior information, protein–protein interaction (PPI) network scores, directly into the LASSO regularization penalty. Furthermore, we introduce a tunable parameter Φ that controls how strongly biological prior knowledge steers feature selection.
# Install
devtools::install_github("dmhenke/BioPrimeLASSO")
# Core usage
scores <- get_scores(gene = "EGFR", network = ppi)
results <- bplasso(scale(X), y, scores,
n_folds = 10,
phi_range = seq(0, 1, length = 30))📄 Published: Henke et al. (2025). Bio-primed machine learning to enhance discovery of relevant biomarkers. npj Precision Oncology, 9, 39. → Read the paper
| Domain | Details |
|---|---|
| Regularized Regression | Biologically-informed LASSO, adaptive penalties, penalty weighting with network priors |
| Functional Genomics | DepMap CRISPR/RNAi dependency screens, DEMETER2, Chronos |
| Network Biology | PPI integration (STRING DB), co-dependency analysis, synthetic lethality |
| Precision Oncology | Biomarker discovery, copy number variation, gene dependency, drug sensitivity prediction |
| High-Dimensional Statistics | Feature selection in p >> n regimes, collinearity, reproducible analysis pipelines |
Learn to appreciate the mountain on which you're already standing