LoginSign Up
← Back to all sparks
D

driveR

INFRA · APIS
Velocity0.0

Prioritizing Cancer Driver Genes Using Genomics Data

github.com/egeulgen/driver ↗

A cancer driver prioritization package that ships rarely and mostly to stay installable

cancer-genomicsbioinformaticsr-packagedriver-genesmachine-learning
◆Current state
driveR prioritizes cancer driver genes from somatic variant and copy number data, combining coding impact scores, noncoding impact, copy number alteration scores and hotspot annotations into a multi-task learning classification model. Version 0.5.0 added gene-level SCNA data frames as an accepted input to create_features_df(), with an example table shipped alongside, widening the entry point beyond the segment-level format. The same release moved org.Hs.eg.db and both hg19 and hg38 TxDb annotation packages from Imports to Suggests under new CRAN policy, with dependent functions now raising an error when they are absent rather than silently degrading.
◆Where it's heading
Releases are infrequent and split cleanly between capability and correction. GRCh38 support arrived in 0.4.0 and cancer-type-specific thresholds were refreshed in 0.3.0, while the 0.2.x pair fixed scoring errors serious enough to require retraining: a column name mismatch meant the SCNA score was not being computed at all, and MCR table coordinates needed converting from hg18 to hg19. Both times the bundled classification model and thresholds were rebuilt as a consequence. Since 0.4.0 the changes have been input handling and packaging rather than method.
◆Prediction
The move of the annotation databases to Suggests suggests a leaner install is the current priority; the entries give no indication of planned model or scoring changes.

◆Recent moves

  1. 7mo ago

    Gene-level copy number input accepted, annotation packages made optional

    Accepts gene-level SCNA data frames in create_features_df(), which removes a preprocessing step for users whose copy number data already sits at gene resolution, and ships an example table to document the format. The annotation database move to Suggests follows CRAN policy and cuts what a plain install pulls in, at the cost of an error rather than a fallback when they are missing.

    View source ↗
  2. 3y ago

    CRAN documentation error fixed

    A single fix for a package documentation issue flagged by CRAN checks. The only release in a three-year stretch, and it contains nothing user-facing.

    View source ↗
  3. 4y ago

    GRCh38 genome build supported

    Adds GRCh38 support, letting the package work with the reference build most current sequencing pipelines produce rather than requiring liftover to hg19. A one-line release note for what is the most consequential change in the package's recent history.

    View source ↗
  4. 4y ago

    Cancer-type-specific thresholds updated

    Refreshes the cancer-type-specific thresholds that determine which genes are called drivers, and updates citations to the published method article. Threshold changes alter which genes a given dataset returns, so this is more than housekeeping despite the short note.

    View source ↗
  5. 5y ago

    MCR coordinates converted to hg19 and the model retrained

    Fixes MCR table coordinates that were still on hg18 and rebuilds the multi-task learning model on the corrected data. The second consecutive release where a data-level error propagated into the shipped model, which is the risk of bundling a trained classifier with its reference tables.

    View source ↗
  6. 5y ago

    Copy number score was never being computed, model rebuilt

    A column name assigned as transcript_overlap_percent instead of MCR_overlap_percent meant create_SCNA_score_df() produced no SCNA score at all, and the classification model and thresholds had to be retrained once it was fixed. Also generalizes hotspot gene determination to accept occurrence annotations from different COSMIC versions.

    View source ↗