## Changes * Updated versions but no support for scipy 1.16.0 yet because of https://github.com/statsmodels/statsmodels/issues/9584#issue-3166348079 * Dataset.py: FeatureDataset from_csv now extracts meta info, to_csv saves it * Removed hard-coded cell line/drug identifiers in visualization/utils.py ## 🐛 Bug Fixes * various fixes (#235) @PascalIversen: * Fixed inverse transform in cross_study_prediction * Inverse transforming in train_and_predict for train and early stopping as well now * Removed nested final_model directory * Copying drug features before supplying it to train/predict functions, too, now. Forgot that previously

## Changes * Bumped Dockerfile to Python 3.13, too, and added unzip for Nextflow

## Small Changes * Defaults for visualization result and data directory * Fixed bug that CD diagram is overwritten * Updated documentation * New package versions

## Changes * New support for Leave-tissue-out cross-validation (#192) @PascalIversen * Implemented ProteomicsRF (#183) @JudithBernett * Included NaiveTissueMeanPredictor * New safety scan * [DEPENDABOT]: Bump codecov/codecov-action from 5.4.0 to 5.4.2 (#196) @[dependabot[bot]](https://github.com/apps/dependabot) * [DEPENDABOT]: Bump crazy-max/ghaction-github-labeler from 5.2.0 to 5.3.0 (#179) @[dependabot[bot]](https://github.com/apps/dependabot) * [DEPENDABOT]: Bump actions/cache from 4.2.2 to 4.2.3 (#172) @[dependabot[bot]](https://github.com/apps/dependabot) ## 🐛 Bug Fixes * fix LTO for low tissue validations sets (#202) @PascalIversen * Lots of restructuring in visualization, kick out partial correlation, instead normalized metrics (#81) @JudithBernett * fix measure being not picked for datasets (#186) @PascalIversen * Multiomics input fix for MultiOmicsNN (#201) @JudithBernett * Fixed an error for early splitting for when there are less than 4 groups in validation ## 🧰 Maintenance * Update README.md (#198) @PascalIversen * Link Hosted Report in Readme (#187) @PascalIversen

## Changes * forced NaiveMeanEffectsPredictor (#169) @PascalIversen ## 🐛 Bug Fixes * fix logic when no baseline is selected (#176) @PascalIversen * fixed randomization. (#180) @PascalIversen * Minor bug fixes for DIPK, MOLIR, and SuperFELTR (#161) @JudithBernett

## Changes * forced NaiveMeanEffectsPredictor (#169) @PascalIversen * updated requirements * test restructuring: now baseline tests, single drug model tests, neural network tests for non-single drug models, adapted tests for the bugs * MultiOmicsNN: removed unused self.methylation_features ## 🐛 Bug Fixes * Minor bug fixes for DIPK, MOLIR, and SuperFELTR (#161) @JudithBernett * DIPK fix for batch size 1 * MOLIR/SuperFELTR: removed VarianceThreshold selection. Now always selecting min(1000, n_features) most variable features from the dataset * MOLIR/SuperFELTR fix for if self.model is None * MOLIR/SuperFELTR: moved duplicated code to utils

## Changes * New package versions ## 🐛 Bug Fixes * Metrics fix (#159) @JudithBernett : If there were NaNs in the predicted values, the metrics could not be computed -> also adapted test

Many things were going wrong for cross study prediction -> fixed ## Changes * Instead of Toy_Data, we now have 2 datasets: TOYv1 (subsetted from CTRPv2) and TOYv2 (subsetted from GDSC2) for testing cross study * We needed lists that contain the intersection of all OMICs features across datasets * rename Toy_Data and add TOYv2 for testing cross study predictions (#153) @PascalIversen * load_and_reduce_gene_features -> load_and_select_gene_features * New versions of packages * [DEPENDABOT]: Bump actions/cache from 4.2.1 to 4.2.2 (#146) @[dependabot[bot]](https://github.com/apps/dependabot) * [DEPENDABOT]: Bump codecov/codecov-action from 5.3.1 to 5.4.0 (#145) @[dependabot[bot]](https://github.com/apps/dependabot) ## 🐛 Bug Fixes * Consolidation fix (#156) @JudithBernett * The features have to have the same order for all datasets -> now happens in load_and_select_gene_features: fix the order of genes across datasets when selecting a gene/methylation island subset of features (#154) @PascalIversen * Bug for MOLIR, SuperFELTR: meta-info of feature dataset was not subsetted for variancethreshold filtering * Bug for MOLIR, SuperFELTR: features used in training have to be remembered such that the cross-study prediction dataset can be subsetted * Bug for MultiOMICs RF: methylation features have to occur in both datasets -> now uses list * Bug for pipeline: consolidate single drug model predictions was receiving cross study datasets instead of the dataset names * DIPK fix: gene expression data has to be encoded again for cross study prediction updated tests