Add multiomics cancer classification in the tutorials.

.gitignore

@@ -180,3 +180,6 @@ _build/
 .vscode/
 
 _build
+
+.DS_Store
+tutorials/multiomics-cancer-classification/outputs

tutorials/multiomics-cancer-classification/config.py

@@ -0,0 +1,61 @@
+"""
+Default configurations according to the MOGONET method described in 'MOGONET integrates multi-omics data using
+graph convolutional networks allowing patient classification and biomarker identification'
+- Wang, T., Shao, W., Huang, Z., Tang, H., Zhang, J., Ding, Z., Huang, K. (2021).
+
+https://github.com/txWang/MOGONET/blob/main/main_mogonet.py
+"""
+
+from yacs.config import CfgNode
+
+# ---------------------------------------------------------
+# Config definition
+# ---------------------------------------------------------
+
+_C = CfgNode()
+
+# ---------------------------------------------------------
+# Dataset
+# ---------------------------------------------------------
+_C.DATASET = CfgNode()
+_C.DATASET.ROOT = "dataset/"
+_C.DATASET.NAME = "TCGA_BRCA"
+_C.DATASET.URL = "https://github.com/pykale/data/raw/main/multiomics/TCGA_BRCA.zip"
+_C.DATASET.RANDOM_SPLIT = False
+_C.DATASET.NUM_MODALITIES = 3  # Number of omics modalities in the dataset
+_C.DATASET.NUM_CLASSES = 5
+
+# ---------------------------------------------------------
+# Solver
+# ---------------------------------------------------------
+_C.SOLVER = CfgNode()
+_C.SOLVER.SEED = 2023
+_C.SOLVER.MAX_EPOCHS_PRETRAIN = 500
+_C.SOLVER.MAX_EPOCHS = 2500
+
+# -----------------------------------------------------------------------------
+# Model (MOGONET) configs
+# -----------------------------------------------------------------------------
+_C.MODEL = CfgNode()
+_C.MODEL.EDGE_PER_NODE = (
+    10  # Predefined number of edges per nodes in computing adjacency matrix
+)
+_C.MODEL.EQUAL_WEIGHT = False
+_C.MODEL.GCN_LR_PRETRAIN = 1e-3
+_C.MODEL.GCN_LR = 5e-4
+_C.MODEL.GCN_DROPOUT_RATE = 0.5
+_C.MODEL.GCN_HIDDEN_DIM = [400, 400, 200]
+
+# The View Correlation Discovery Network (VCDN) to learn the higher-level intra-view and cross-view correlations
+# in the label space. See the MOGONET paper for more information.
+_C.MODEL.VCDN_LR = 1e-3
+
+# ---------------------------------------------------------
+# Misc options
+# ---------------------------------------------------------
+_C.OUTPUT = CfgNode()
+_C.OUTPUT.OUT_DIR = "./outputs"
+
+
+def get_cfg_defaults():
+    return _C.clone()

tutorials/multiomics-cancer-classification/experiments/MOGONET_ROSMAP_quick_test.yaml

@@ -0,0 +1,13 @@
+DATASET:
+  NAME: "ROSMAP"
+  URL: "https://github.com/pykale/data/raw/main/multiomics/ROSMAP.zip"
+  NUM_MODALITIES: 3
+  NUM_CLASSES: 2
+
+SOLVER:
+  MAX_EPOCHS_PRETRAIN: 5 # For quick testing
+  MAX_EPOCHS: 10  # For quick testing
+
+MODEL:
+  EDGE_PER_NODE: 2
+  GCN_HIDDEN_DIM: [200, 200, 100]

tutorials/multiomics-cancer-classification/experiments/base.yaml

@@ -0,0 +1,13 @@
+DATASET:
+  NAME: "TCGA_BRCA"
+  URL: "https://github.com/pykale/data/raw/main/multiomics/TCGA_BRCA.zip"
+  NUM_MODALITIES: 3
+  NUM_CLASSES: 5
+
+SOLVER:
+  MAX_EPOCHS_PRETRAIN: 5 # For quick testing
+  MAX_EPOCHS: 100  # For quick testing
+
+MODEL:
+  EDGE_PER_NODE: 10
+  GCN_HIDDEN_DIM: [400, 400, 200]

tutorials/multiomics-cancer-classification/model.py

@@ -0,0 +1,102 @@
+from typing import List, Optional
+
+from torch.nn import CrossEntropyLoss
+from yacs.config import CfgNode
+
+from kale.embed.mogonet import MogonetGCN
+from kale.loaddata.multiomics_datasets import SparseMultiomicsDataset
+from kale.pipeline.multiomics_trainer import MultiomicsTrainer
+from kale.predict.decode import LinearClassifier, VCDN
+
+
+class MogonetModel:
+    r"""Setup the MOGONET model via the config file.
+
+    Args:
+        cfg (CfgNode): A YACS config object.
+        dataset (SparseMultiomicsDataset): The input dataset created in form of :class:`~torch_geometric.data.Dataset`.
+    """
+
+    def __init__(self, cfg: CfgNode, dataset: SparseMultiomicsDataset) -> None:
+        self.cfg = cfg
+        self.dataset = dataset
+        self.unimodal_encoder: List[MogonetGCN] = []
+        self.unimodal_decoder: List[LinearClassifier] = []
+        self.multimodal_decoder: Optional[VCDN] = None
+        self.loss_function = CrossEntropyLoss(reduction="none")
+        self._create_model()
+
+    def _create_model(self) -> None:
+        """Create the MOGONET model via the config file."""
+        num_modalities = self.cfg.DATASET.NUM_MODALITIES
+        num_classes = self.cfg.DATASET.NUM_CLASSES
+        gcn_dropout_rate = self.cfg.MODEL.GCN_DROPOUT_RATE
+        gcn_hidden_dim = self.cfg.MODEL.GCN_HIDDEN_DIM
+        vcdn_hidden_dim = pow(num_classes, num_modalities)
+
+        for modality in range(num_modalities):
+            self.unimodal_encoder.append(
+                MogonetGCN(
+                    in_channels=self.dataset.get(modality).num_features,
+                    hidden_channels=gcn_hidden_dim,
+                    dropout=gcn_dropout_rate,
+                )
+            )
+
+            self.unimodal_decoder.append(
+                LinearClassifier(in_dim=gcn_hidden_dim[-1], out_dim=num_classes)
+            )
+
+        if num_modalities >= 2:
+            self.multimodal_decoder = VCDN(
+                num_modalities=num_modalities,
+                num_classes=num_classes,
+                hidden_dim=vcdn_hidden_dim,
+            )
+
+    def get_model(self, pretrain: bool = False) -> MultiomicsTrainer:
+        """Return the prepared MOGONET model based on user preference.
+
+        Args:
+            pretrain (bool, optional): Whether to return the pretrain model. (default: ``False``)
+
+        Returns:
+            MultiomicsTrainer: The prepared MOGONET model.
+        """
+        num_modalities = self.cfg.DATASET.NUM_MODALITIES
+        num_classes = self.cfg.DATASET.NUM_CLASSES
+        gcn_lr_pretrain = self.cfg.MODEL.GCN_LR_PRETRAIN
+        gcn_lr = self.cfg.MODEL.GCN_LR
+        vcdn_lr = self.cfg.MODEL.VCDN_LR
+
+        if pretrain:
+            multimodal_model = None
+            train_multimodal_decoder = False
+            gcn_lr = gcn_lr_pretrain
+        else:
+            multimodal_model = self.multimodal_decoder
+            train_multimodal_decoder = True
+            gcn_lr = gcn_lr
+
+        model = MultiomicsTrainer(
+            dataset=self.dataset,
+            num_modalities=num_modalities,
+            num_classes=num_classes,
+            unimodal_encoder=self.unimodal_encoder,
+            unimodal_decoder=self.unimodal_decoder,
+            loss_fn=self.loss_function,
+            multimodal_decoder=multimodal_model,
+            train_multimodal_decoder=train_multimodal_decoder,
+            gcn_lr=gcn_lr,
+            vcdn_lr=vcdn_lr,
+        )
+
+        return model
+
+    def __str__(self) -> str:
+        r"""Returns a string representation of the model object.
+
+        Returns:
+            str: The string representation of the model object.
+        """
+        return self.get_model().__str__()