Changes of last 4 months (code, not data)

.gitignore

@@ -128,3 +128,27 @@ dmypy.json
 # Pyre type checker
 .pyre/
 .DS_Store
+
+# Checkpoints (too heavy to upload them all)
+checkpoints/agnostic_all_modes/seed_13/
+checkpoints/agnostic_all_modes/seed_23/
+checkpoints/agnostic_all_modes/seed_398/
+checkpoints/agnostic_all_modes/seed_447/
+checkpoints/contact_maps/seed_13/
+checkpoints/contact_maps/seed_23/
+checkpoints/contact_maps/seed_398/
+checkpoints/contact_maps/seed_447/
+checkpoints/contact_maps/seed_917/
+checkpoints/full_ags_all_modes/seed_13/
+checkpoints/full_ags_all_modes/seed_23/
+checkpoints/full_ags_all_modes/seed_398/
+checkpoints/full_ags_all_modes/seed_447/
+checkpoints/full_ags_all_modes/seed_917/
+
+# Data not used
+data/contact_maps_explorer/
+data/contact_maps_with_antigen_explorer/
+notebooks/biological_properties/
+
+# Raw notebooks
+notebooks/explainability_raw.ipynb

AUTHORS.md

@@ -8,5 +8,5 @@ Core developer:
 
 Supervisors:
 
-- Barbara Bravi [[Website](https://www.imperial.ac.uk/people/b.bravi21)|[Github](https://github.com/bravib)|[Email](mailto:b.bravi21@imperial.ac.uk?subject=[GitHub]%20ANTIPASTI)]
-- Mauricio Barahona [[Website](https://www.imperial.ac.uk/people/m.barahona/)|[Github](https://github.com/mauriciobarahona)|[Email](mailto:m.barahona@imperial.ac.uk?subject=[GitHub]%20ANTIPASTI)]
+- Mauricio Barahona [[Website](https://www.imperial.ac.uk/people/m.barahona/)|[Github](https://github.com/mauriciobarahona)|[Email](mailto:m.barahona@imperial.ac.uk?subject=[GitHub]%20ANTIPASTI)]
+- Barbara Bravi [[Website](https://www.imperial.ac.uk/people/b.bravi21)|[Github](https://github.com/bravib)|[Email](mailto:b.bravi21@imperial.ac.uk?subject=[GitHub]%20ANTIPASTI)]

README.md

@@ -43,9 +43,11 @@ python setup.py install --user
 
 ANTIPASTI requires the following Python packages: 
 * `adabelief-pytorch`
+* `biopython`
 * `matplotlib`
 * `numpy`
 * `opencv-python`
+* `optuna`
 * `pandas`
 * `scikit-learn`
 * `torch`

antipasti-env.yml

@@ -16,5 +16,7 @@ dependencies:
   - pip:
     - adabelief-pytorch>=0.2.1
     - beautifulsoup4>=4.12.2
+    - biopython>=1.79
     - opencv-python>=4.7.0.68
+    - optuna>=3.1.1
     - torch>=1.13.1

antipasti/model/model.py

@@ -22,6 +22,10 @@ class ANTIPASTI(Module):
         Size of the max pooling operation.
     input_shape: int
         Shape of the normal mode correlation maps.
+    l1_lambda: float
+        Weight of L1 regularisation.
+    mode: str
+        To use the full model, provide ``full``. Otherwise, ANTIPASTI corresponds to a linear classifier.
 
     """
     def __init__(
@@ -31,27 +35,27 @@ def __init__(
             pooling_size=1,
             input_shape=281,
             l1_lambda=0.002,
+            mode='full',
     ):
         super(ANTIPASTI, self).__init__()
         self.n_filters = n_filters
         self.filter_size = filter_size
         self.pooling_size = pooling_size
         self.input_shape = input_shape
-        self.fully_connected_input = n_filters * ((input_shape-filter_size+1)//pooling_size) ** 2
-        self.conv1 = Conv2d(1, n_filters, filter_size)
-        #self.conv1_bn = torch.nn.BatchNorm2d(n_filters)
-        #torch.nn.init.normal_(self.conv1.weight, mean=0.0, std=1/input_shape)
-        #torch.nn.init.constant_(self.conv1.bias, 0)
-        self.pool = MaxPool2d(pooling_size, pooling_size)
-        self.dropit = Dropout(p=0.05)
-        self.relu = ReLU()
+        self.mode = mode
+        if self.mode == 'full':
+            self.fully_connected_input = n_filters * ((input_shape-filter_size+1)//pooling_size) ** 2
+            self.conv1 = Conv2d(1, n_filters, filter_size)
+            self.pool = MaxPool2d(pooling_size, pooling_size)
+            #self.dropit = Dropout(p=0.05)
+            self.relu = ReLU()
+        else:
+            self.fully_connected_input = self.input_shape ** 2
         self.fc1 = Linear(self.fully_connected_input, 1, bias=False)
-        #self.fc1_bn = torch.nn.BatchNorm1d(self.fully_connected_input)
-        #torch.nn.init.normal_(self.fc1.weight, mean=0.0, std=1/np.sqrt(self.fully_connected_input))
-        #self.fc2 = Linear(self.fully_connected_input, 1, bias=False)
+        #self.fc2 = Linear(4, 1, bias=False)
         self.l1_lambda = l1_lambda
 
-    def forward(self, input):
+    def forward(self, x):
         r"""Model's forward pass.
 
         Returns
@@ -62,24 +66,16 @@ def forward(self, input):
             Filters before the fully-connected layer.
             
         """
-        #if torch.numel(torch.nonzero(input[0,0,-80:,-80:])) == 0:
-        #    x = self.conv2(input) + torch.transpose(self.conv2(input), 2, 3)
-        #else:
-        #    x = self.conv1(input) + torch.transpose(self.conv1(input), 2, 3)
-        x = self.conv1(input) + torch.transpose(self.conv1(input), 2, 3)
-        #x = self.conv1_bn(x)
-        x = self.relu(x)
-        x = self.pool(x)
-        inter = x 
+        inter = x
+        if self.mode == 'full':
+            x = self.conv1(x) + torch.transpose(self.conv1(x), 2, 3)
+            x = self.relu(x)
+            inter = x = self.pool(x)
         x = x.view(x.size(0), -1)
-        x = self.dropit(x)
-        #if torch.numel(torch.nonzero(input[0,0,-80:,-80:])) == 0:
-        #    x = self.fc2(x)
-        #    print('nano')
-        #else:
-        #    x = self.fc1(x)
-        #    print('paired')
+        #if self.mode == 'full':
+        #    x = self.dropit(x)
         x = self.fc1(x)
+        #x = self.fc2(x)
 
         return x.float(), inter