release MixTcpred_v1.0

GfellerLab · Mar 11, 2024 · 950313a · 950313a
1 parent 2b64bac
commit 950313a
Show file tree

Hide file tree

Showing 3 changed files with 26 additions and 22 deletions.
diff --git a/MixTCRpred.py b/MixTCRpred.py
@@ -8,7 +8,8 @@
 import configparser
 import pandas as pd
 
-path_pretrained_models = './pretrained_models'
+path_pretrained_models = '/home/giancarlo/Documents/lab_work/MixTCRpred/pretrained_models'
+#path_pretrained_models = './pretrained_models'
 
 if __name__ == '__main__':
 

diff --git a/src/imgt_ref_seq/get_CDR12_fromVgene.py b/src/imgt_ref_seq/get_CDR12_fromVgene.py
@@ -22,6 +22,7 @@
     name = seq_record.id
     vgene = name.split("|")[1]
     seq = seq_record.seq
+    break
     cdr1 = str(seq[26:38]).replace("-","")
     cdr2 = str(seq[55:65]).replace("-","")
     all_cdr1.append(cdr1)
@@ -62,6 +63,7 @@
     name = seq_record.id
     vgene = name.split("|")[1]
     seq = seq_record.seq
+    print(len(seq))
     cdr1 = str(seq[26:38]).replace("-","")
     cdr2 = str(seq[55:65]).replace("-","")
     all_cdr1.append(cdr1)

diff --git a/src/models.py b/src/models.py
@@ -37,7 +37,7 @@ def __init__(self, vocab_size, embedding_dim,  hidden_dim, num_heads, num_layers
         self.scale = torch.sqrt(torch.FloatTensor([self.embedding_dim]))
         self.embedding_pos_epi = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[0])
         self.embedding_pos_TRA = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[1] + 2*self.padding[-1])
-        self.embedding_pos_TRB = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[1] + 2*self.padding[-1])
+        self.embedding_pos_TRB = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[2] + 2*self.padding[-1])
         # Transformer - Encoder
         self.transformer_encoder = TransformerEncoder(num_layers=num_layers, input_dim=embedding_dim, dim_feedforward=hidden_dim, num_heads=num_heads, dropout=dropout)
         ### Output classifier
@@ -131,23 +131,23 @@ def training_step(self, batch, batch_idx):
         preds = self.forward(inp_data, mask = True)
         loss = self.loss_function(preds, labels)
         self.log('train_loss', loss)
-        #compute auc
-        fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
-        AUC = metrics.auc(fpr, tpr)
-        #print("AUC_train:{0}".format(AUC))
-        self.log('train_auc', AUC)
+        ##compute auc
+        #fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
+        #AUC = metrics.auc(fpr, tpr)
+        ##print("AUC_train:{0}".format(AUC))
+        #self.log('train_auc', AUC)
         return loss
     def validation_step(self, batch, batch_idx):
         inp_data = batch[1]
         labels = batch[2]
         preds = self.forward(inp_data, mask = True)
         loss = self.loss_function(preds, labels)
         self.log('val_loss', loss)
-        #compute auc
-        fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
-        AUC = metrics.auc(fpr, tpr)
-        #print("AUC_train:{0}".format(AUC))
-        self.log('val_auc', AUC)
+        ##compute auc
+        #fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
+        #AUC = metrics.auc(fpr, tpr)
+        ##print("AUC_train:{0}".format(AUC))
+        #self.log('val_auc', AUC)
         return loss
     def test_step(self, batch, batch_idx):
         test_seq = batch[0]
@@ -156,6 +156,7 @@ def test_step(self, batch, batch_idx):
         preds = self.forward(inp_data, mask = True)
         loss = self.loss_function(preds, labels)
         self.prob.extend(preds.data[:,1].cpu().numpy())
+        #self.prob.extend(preds.data.cpu().numpy())
         self.test_tp.extend(labels.cpu().numpy())
         self.test_seq.extend(test_seq)
 
@@ -188,7 +189,7 @@ def __init__(self, vocab_size, embedding_dim,  hidden_dim, num_heads, num_layers
         self.embedding_pos_cdr12= PositionWiseEmbedding(self.vocab_size, self.embedding_dim, 4*self.padding[-1])
         ########## TEST ############3
         self.embedding_pos_TRA = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[1] + 2*self.padding[-1])
-        self.embedding_pos_TRB = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[1] + 2*self.padding[-1])
+        self.embedding_pos_TRB = PositionWiseEmbedding(self.vocab_size, self.embedding_dim, self.padding[2] + 2*self.padding[-1])
         # Transformer - Encoder
         self.transformer_encoder = TransformerEncoder(num_layers=num_layers,
                                               input_dim=embedding_dim,
@@ -344,22 +345,22 @@ def training_step(self, batch, batch_idx):
         loss = self.loss_function(preds, labels)
         self.log('train_loss', loss)
         #compute auc
-        fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
-        AUC = metrics.auc(fpr, tpr)
-        #print("AUC_train:{0}".format(AUC))
-        self.log('train_auc', AUC)
+        #fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
+        #AUC = metrics.auc(fpr, tpr)
+        ##print("AUC_train:{0}".format(AUC))
+        #self.log('train_auc', AUC)
         return loss
     def validation_step(self, batch, batch_idx):
         inp_data = batch[1]
         labels = batch[2]
         preds = self.forward(inp_data, mask = True)
         loss = self.loss_function(preds, labels)
         self.log('val_loss', loss)
-        #compute auc
-        fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
-        AUC = metrics.auc(fpr, tpr)
-        #print("AUC_train:{0}".format(AUC))
-        self.log('val_auc', AUC)
+        ##compute auc
+        #fpr, tpr, threshold = metrics.roc_curve(labels.cpu().numpy(), preds.data[:,1].cpu().numpy())
+        #AUC = metrics.auc(fpr, tpr)
+        ##print("AUC_train:{0}".format(AUC))
+        #self.log('val_auc', AUC)
         return loss
     def test_step(self, batch, batch_idx):
         test_seq = batch[0]