init

dummy.csv

+Sequences,Classes,Proteins
+IVDDLSALTVLEASELSK,0,0
\ No newline at end of file

main.py

+import numpy as np
+import pandas as pd
+from dlomix.models import DetectabilityModel
+from dlomix.constants import CLASSES_LABELS, alphabet, aa_to_int_dict
+from dlomix.data import DetectabilityDataset
+
+def fasta_like_to_data(path):
+    file = open(path, "r")
+    content = file.readlines()
+    file.close()
+    return content
+
+def strip_lines(s):
+    s = s.strip('\n')
+    s = s.split(' ')[1]
+    return s
+
+def main(input_data_path):
+    print('Reading file')
+    file = open(input_data_path, "r")
+    content = file.readlines()
+    file.close()
+    print('Converting file')
+    data = pd.DataFrame(content,columns=['Sequences'])
+    data = data[~data['Sequences'].str.contains(">")]
+    data['Sequences']=data['Sequences'].map(strip_lines)
+
+    data['Classes']=[0]*data.shape[0]
+    data['Proteins']=[0]*data.shape[0]
+    data.to_csv('temp.csv',index=False)
+
+    print('Initialising model')
+    ## Model init
+    total_num_classes = len(CLASSES_LABELS)
+    input_dimension = len(alphabet)
+    num_cells = 64
+
+    model = DetectabilityModel(num_units = num_cells, num_clases = total_num_classes)
+
+    ## Loading model weights
+    model.built = True
+    model_save_path = 'pretrained_model/original_detectability_fine_tuned_model_FINAL'
+
+    model.load_weights(model_save_path)
+    max_pep_length = 40
+    ## Has no impact for prediction
+    batch_size = 128
+
+    print('Initialising dataset')
+    ## Data init
+    detectability_data = DetectabilityDataset(data_source= 'dummy.csv',
+                                              val_data_source= 'temp.csv',
+                                              data_format='csv',
+                                              max_seq_len=max_pep_length,
+                                              label_column="Classes",
+                                              sequence_column="Sequences",
+                                              dataset_columns_to_keep=['Proteins'],
+                                              batch_size=batch_size,
+                                              with_termini=False,
+                                              alphabet=aa_to_int_dict)
+    val_data = detectability_data.tensor_val_data
+    seq = detectability_data["val"]["_parsed_sequence"]
+    seq = list(map(lambda x: "".join(x),seq))
+
+    print('Applying model')
+    ## Applying model
+    predictions = model.predict(val_data)
+
+    print('Formatting output')
+    ## Formatting output
+    label_multi = np.argmax(predictions, axis=1)
+    label_multi = list(map(lambda x : CLASSES_LABELS[x],label_multi))
+    result = pd.DataFrame({'Sequences':seq,'Probability no flyer':predictions[:,0],'Probability low flyer':predictions[:,1],
+                           'Probability medium flyer':predictions[:,2],'Probability high flyer':predictions[:,3],
+                           'Predicted class':label_multi})
+
+    print('Saving output')
+    result.to_csv(input_data_path.strip('.txt')+'_detectability.csv',index=False) #Output du model dans le fichier 'prediction_detectability.csv'
+
+    new_file = open(input_data_path.strip('.txt')+'_filtered.txt','w')
+
+    print('Writting final file')
+    i,j=0,1
+    for l in content :
+        if '>' in l :
+            j=1
+            new_file.write(l)
+        elif result['Predicted class'][i]!='Non-Flyer':
+            l_s = l.split(' ')
+            new_file.write(str(j)+' '+l_s[1])
+            i+=1
+            j+=1
+        else :
+            i+=1
+    new_file.close()
+
+if __name__ == '__main__':
+    main('241211_FASTA_RP_GroEL_GroES_Tuf_assemble_peptides_list.txt')
\ No newline at end of file

pretrained_model/checkpoint

+model_checkpoint_path: "fine_tuned_weights_attention_model_FINAL_NON"
+all_model_checkpoint_paths: "fine_tuned_weights_attention_model_FINAL_NON"