Fix PaccMann integration

docs/drevalpy.models.PaccMann.rst

@@ -0,0 +1,50 @@
+PaccMann
+=============================
+
+PaccMann Model
+----------------------------------
+
+.. automodule:: drevalpy.models.PaccMann.paccmann
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+PaccMannV2 Network
+----------------------------------
+
+.. automodule:: drevalpy.models.PaccMann.paccmann_network_v2
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+Hyperparameter utils
+----------------------------------
+
+.. automodule:: drevalpy.models.PaccMann.utils.hyperparams
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+Layers
+----------------------------------
+
+.. automodule:: drevalpy.models.PaccMann.utils.layers
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+Loss functions
+----------------------------------
+
+.. automodule:: drevalpy.models.PaccMann.utils.loss_functions
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+Model utils
+----------------------------------
+
+.. automodule:: drevalpy.models.PaccMann.utils.utils
+   :members:
+   :undoc-members:
+   :show-inheritance:

docs/drevalpy.models.rst

@@ -27,6 +27,7 @@ Implemented models
    drevalpy.models.DIPK
    drevalpy.models.DrugGNN
    drevalpy.models.MOLIR
+   drevalpy.models.PaccMann
    drevalpy.models.PharmaFormer
    drevalpy.models.Precily
    drevalpy.models.SRMF

docs/installation.rst

@@ -65,6 +65,10 @@ default ``pip install drevalpy``. They are provided as optional `extras`:
    * - ``xgboost``
      - The ``MultiViewXGBoost`` baseline model
      - ``xgboost``
+   * - ``paccmann``
+     - SMILES augmentation for the ``PaccMann`` model. Without it, PaccMann trains on the
+       unaugmented SMILES and warns.
+     - ``rdkit``
    * - ``multiprocessing``
      - Parallelized cross-validation / tuning via Ray
      - ``ray`` (and ``pydantic``, usually already present)

docs/usage.rst

@@ -302,6 +302,8 @@ See the sklearn model :ref:`flexible-inputs` or the SimpleNeuralNetwork :ref:`fl
 +---------------------------------+----------------------------+--------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 | Precily                         | Published Model            | Multi-Drug Model                     | `Precily <https://doi.org/10.1038/s41467-022-33291-z>`_ from Chawla et al. Uses GSVA pathway-activity scores with SMILESVec drug embeddings. Features are concatenated and passed through multiple linear layers with ReLU and Dropout.                                                                                                                                                                                                                                                                                                                                                    |
 +---------------------------------+----------------------------+--------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
+| PaccMann                        | Published Model            | Multi-Drug Model                     | `PaccMann <https://doi.org/10.1021/acs.molpharmaceut.9b00520>`_ from Manica et al. Embeds tokenized drug SMILES and encodes them with multi-scale convolutional layers, while cell line gene expression of a curated gene panel serves as biological context. Contextual attention layers connect the gene and molecule representations, which are concatenated and passed through stacked dense layers to predict the response.                                                                                                                                                           |
++---------------------------------+----------------------------+--------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 
 
 Available Datasets

drevalpy/models/PaccMann/__init__.py

@@ -0,0 +1,5 @@
+"""Module for the Paccmann model."""
+
+from .paccmann import PaccMann
+
+__all__ = ["PaccMann"]

drevalpy/models/PaccMann/hyperparameters.yaml

@@ -0,0 +1,59 @@
+PaccMann:
+  gene_list:
+    - gene_list_paccmann_network_prop
+
+  epochs:
+    - 10
+  batch_size:
+    - 64
+  learning_rate:
+    - 0.001
+  weight_decay:
+    - 0.0
+
+  smiles_embedding_size:
+    - 8
+
+  filters:
+    - [16, 16, 16]
+
+  molecule_heads:
+    - [2, 2, 2, 2]
+
+  gene_heads:
+    - [2, 2, 2, 2]
+
+  smiles_padding_length:
+    - 512
+
+  # Train each drug on several equivalent SMILES strings. Requires rdkit; without it
+  # training falls back to the unaugmented SMILES and warns.
+  augment_smiles:
+    - true
+
+  dropout:
+    - 0.5
+
+  batch_norm:
+    - true
+
+  activation_fn:
+    - relu
+
+  loss_fn:
+    - mse
+
+  smiles_attention_size:
+    - 64
+
+  gene_attention_size:
+    - 1
+
+  molecule_temperature:
+    - 1.0
+
+  gene_temperature:
+    - 1.0
+
+  stacked_dense_hidden_sizes:
+    - [512, 256]