Skip to content

Drugresponseeval 1.4.0: Dazzling Dugtrio

Latest

Choose a tag to compare

@JudithBernett JudithBernett released this 07 Nov 06:32
f16f63a

v1.4.0: Dazzling Dugtrio - 5.11.2026

Added

Changed

  • Single-Drug Models no longer submit one task per drug, hyperparameter combination, and CV split. All drugs are trained within one task, with task.cpus drugs in parallel. All Single-Drug Models run on the CPU, also with the gpu profile. The best hyperparameters are still selected per drug. These tasks get 36 CPUs and 10 h instead of 12 CPUs and 16 h (memory as process_high), by @JudithBernett in #114
  • Moved all params. out of the subworkflows and the DRUGRESPONSEEVAL workflow into NFCORE_DRUGRESPONSEEVAL in main.nf, by @JudithBernett in #120
  • The nf-schema plugin version in the AWS full test workflow and the utils_nfschema_plugin test config is now 2.7.3, matching the pipeline, by @JudithBernett in #120
  • More memory for test profile, by @JudithBernett in #121
  • Full test profile: including LPO and LCO now; switched from ElasticNet to RandomForest,SingleDrugRandomForest; by @JudithBernett in #121
  • Nextflow syntax v2 parser for AWS, by @JudithBernett in #122
  • Template update for nf-core/tools v4.2.1 by @nf-core-bot in #112

Removed

  • MAKE_MODEL_CHANNEL and bin/make_model_channel.py: Single-Drug Models are no longer expanded into one model per drug, so the model names are used directly, by @JudithBernett in #114
  • CONSOLIDATE_RESULTS: the predictions of Single-Drug Models are now consolidated over all drugs in the task that produces them, by @JudithBernett in #114

Fixed

  • With --final_model_on_full_data and more than one test mode, the final model of each test mode was also tuned and trained on the final splits of the other test modes. The final splits are now matched by model and test mode, by @JudithBernett in #115
  • The meta folder needs to match the datasets. It was not properly redownloaded from either Zenodo or GitHub when switching between the Toy datasets and the real ones. By @JudithBernett in #111
  • Processes reading the meta folder from path_data now wait until a redownloaded meta folder has been unzipped, instead of possibly reading the outdated one, by @JudithBernett in #120

Dependencies

Dependency Old version New version
drevalpy 1.6.0 1.7.0
python 3.13.13 3.13.18
numpy 2.5.3 2.5.4
torch 2.14.0 2.15.0
scikit-posthocs 0.17.0 0.17.1

Full Changelog: 1.3.0...1.4.0