add xgdp - #446
Conversation
…ll tests when rdkit not installed
|
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## development #446 +/- ##
===============================================
- Coverage 80.34% 74.97% -5.37%
===============================================
Files 101 122 +21
Lines 8171 10783 +2612
===============================================
+ Hits 6565 8085 +1520
- Misses 1606 2698 +1092 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
|
|
||
| trainer = pl.Trainer( | ||
| # max_epochs=self.hyperparameters.get("epochs", 100), #changed to 10 fro testing | ||
| max_epochs=1, |
There was a problem hiding this comment.
max_epoch=1 is hardcoded here, it doesnt use the epochs from the hpams
| self.gene_expression_scaler: StandardScaler | None = None | ||
| self.gene_expression_normalizer: MinMaxScaler | None = None |
There was a problem hiding this comment.
the scaler and normalizer are declared here but never fit or applied, so the gene expression goes into the cnn unscaled. how does the publication do it?
There was a problem hiding this comment.
Checked the original XGDP repo - gene expression is never scaled/normalized there, only variance basedfiltering is applied before it goes into the CNN. So this is currently dead code. Should I remove it?
6 participants
Adds XGDP. The model combines a GNN branch for drug molecular graphs with a 1D-CNN branch for cell line gene expression, with optional cross-attention.
Adapted from the original repository with the following changes for drevalpy integration:
upstream/developmentand resolved conflictscreate_drug_graphs.pyfeaturizerensure_drug_graphsfixture totests/conftest.pyS614fortorch.load(..., weights_only=False)(required fortorch_geometric.data.Data)