Dev

[emsonder]

Fixes:

• Usage of pre-computed insertion profiles
• Normalization of NA-containing columns (previously all would be "normalized" to NA)

Changes:

• Ensuring order of features explicitly during predictions
• Removing cellular contexts labelled testing/validation from binding pattern computation

[Copilot]

Pull Request Overview

This PR fixes issues with pre-computed insertion profiles and NA-containing column normalization. The changes ensure that feature order is explicitly maintained during predictions and exclude testing/validation datasets from binding pattern computations.

• Fixed usage of pre-computed insertion profiles by properly handling calcProfile parameter
• Added NA-awareness to normalization functions to prevent all-NA results
• Ensured feature ordering matches model expectations during predictions

Reviewed Changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/testthat/test-tfFeatures.R	Added test verifying testing datasets are excluded from binding pattern features
tests/testthat/test-predictTfBinding.R	Added test for feature order matching between model and feature matrix
tests/testthat/test-insertionProfiles.R	Updated message expectations and added test for mismatched pre-computed profiles
tests/testthat/test-getFeatureMatrix.R	Added tests verifying NA preservation in normalization functions
tests/testthat/test-contextTfFeatures.R	Added comprehensive tests for pre-computed profile handling
R/tfFeatures.R	Added informational messages and excluded testing/validation samples from binding patterns
R/predict.R	Replaced column filtering with explicit feature ordering from model
R/helpers.R	Added naRm parameter to margin calculation functions
R/getInsertionProfiles.R	Added validation for pre-computed profiles and improved messaging
R/getFeatureMatrix.R	Added naRm parameter to normalization functions
R/contextTfFeatures.R	Added pre-computed profile handling and motif_id assignment

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The feature order extraction relies on hard-coded line number (keep=8) from the model string representation, which is fragile and may break if the model's string format changes. Consider adding error handling to verify that feature names were successfully extracted, or use a more robust method to access feature names if available in the lightgbm API.

Suggested change

	modText <- model$save_model_to_string()
	featOrder <- unlist(tstrsplit(modText, split="\n", keep=8))
	featOrder <- unlist(tstrsplit(gsub("feature_names=","", featOrder), split=" "))
	featOrder <- model$feature_name()

[emsonder]

model$feature_name() does not seem to exist in the R-API of lightgbm

[plger]

one could still check that the "feature_names=" is present in that row

[emsonder]

agreed, see commit 2ef3949](2ef3949)

[codecov]

Codecov Report

❌ Patch coverage is 96.29630% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
R/getFeatureMatrix.R	85.71%	1 Missing ⚠️
R/predict.R	87.50%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!