Notebook

Preprocessing¶

In [ ]:

user = 'userName'
model_name = 'metal_prediction_CNN'

# Removing invalid samples - already done duplicate_dict = {} rows_to_delete = [] count = 0 for i in range(seqs.shape[0]): if 'X' in seqs[i] \ or 'U' in seqs[i] \ or '3CO' in target[i]\ or '3NI' in target[i] \ or 'FE2'in target[i] \ or 'CU1'in target[i]\ or 'MN3' in target[i] \ or np.isnan(cluster_numbers[i]): rows_to_delete.append(i) count +=1 elif seqs[i] not in duplicate_dict.keys(): duplicate_dict[seqs[i]] = target[i] else: if target[i] != duplicate_dict[seqs[i]]: rows_to_delete.append(i) count +=1 # df = df.drop(df.index[rows_to_delete]) # df.to_parquet('Metal_all_20180601.parquet') seqs = np.delete(seqs, rows_to_delete, 0) target = np.delete(target, rows_to_delete) cluster_numbers = np.delete(cluster_numbers, rows_to_delete)

In [ ]:

import sys
import json

print ("Initializing global variables...", end=' ')
sys.stdout.flush()

# Filepaths
output_file = './logs/results.txt'
hist_path = model_path = fig_path = './logs/'
dict_path = './dictionaries/'

print ("Done")
print ("  Filepath set to ./logs/")

##################################################

print ("Importing modules...", end=' ')
import modules
print ("Done")

##################################################

print ("Reading data from disk...", end=' ')
sys.stdout.flush()

import numpy as np
import pandas as pd

df = pd.read_parquet('./datasets/Metal_all_20180601.parquet')
seqs = np.array(df.sequence)
target = np.array(df.ligandId)
cluster_numbers = np.array(df.clusterNumber90)

for i in range(target.shape[0]):
    target[i] = [label_dict[target[i]]]

print ("Done")

##################################################

print ("Loading dictionaries...", end=' ')
sys.stdout.flush()

# FOFE
vocab_dic_fofe = {}
with open(dict_path + "vocab_dict_fofe", 'r') as fp:
        vocab_dic_fofe = json.load(fp)

print ("Done")

##################################################

print ("Performing cross validation split...", end=' ')
ratio = 0.9
split = int(ratio*len(seqs))
train_seqs, val_seqs = seqs[:split], seqs[split:]
train_label, val_label = target[:split], target[split:]
print ("Done")
print ("  Ratio :", ratio)
print ("  Train_range :", 0, "-", split-1)
print ("  Val_range :", split, "-", len(seqs)-1)

In [ ]:

df.groupby('ligandId').count()

Data Generator¶

FOFE Encoding

In [ ]:

label_dict = {}
with open(dict_path + "metal_dict", 'r') as fp:
        label_dict = json.load(fp)
        
train_args = {'sequences': train_seqs,
              'labels': train_label,
              'translator': vocab_dic_fofe}
val_args = {'sequences': val_seqs,
            'labels': val_label,
            'translator': vocab_dic_fofe}
common_args = {'batch_size': 100,
               'input_shape': (800,),
               'label_shape': (8, ),
               'shuffle': True}

train_gen = modules.FOFEGenerator(**train_args, **common_args)
val_gen = modules.FOFEGenerator(**val_args, **common_args)

Model¶

In [ ]:

# ProtVec:100, One-hot:20, blosum62:20, property:7
dimension = 800
cutoff = 8

import tensorflow as tf
import time
import matplotlib.pyplot as plt
% matplotlib inline
np.random.seed(2017) 
from keras.models import Sequential, Model
from keras.layers.convolutional import Conv2D, MaxPooling2D, Convolution1D, MaxPooling1D, AveragePooling2D
from keras.layers import Activation, Flatten, Dense, Dropout, Reshape, Embedding, Input
from keras.layers.normalization import BatchNormalization
from keras.utils import np_utils
from keras.optimizers import SGD
import numpy as np
import keras
from keras.models import Model, load_model
from keras.optimizers import Adam, SGD, RMSprop
# Visualization
from keras.utils import plot_model

input_shape = (dimension,)

input_0 = Input(shape=input_shape, dtype='float32')
input_0_reshape = Reshape((1,dimension,1), input_shape=(dimension,))(input_0)
conv2d_3 = Conv2D(2, (1, 3), padding='same')(input_0_reshape)
conv2d_5 = Conv2D(2, (1, 5), padding='same')(input_0_reshape)
conv2d_7 = Conv2D(2, (1, 7), padding='same')(input_0_reshape)

x = keras.layers.concatenate([conv2d_3,conv2d_5,conv2d_7])
x = Activation('relu')(x)
x = Flatten()(x)
x = Dense(cutoff, activation='relu')(x)
output_0 = Dense(cutoff, activation='softmax')(x)

model = Model(inputs=input_0, outputs=output_0)                              
# end of the MODEL

sgd = SGD(lr = 0.01, momentum = 0.9, decay = 0, nesterov = False)
model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=['accuracy'])

# model.summary()

In [ ]:

model_args = {'model': model, 
              'generators': [train_gen, val_gen], 
              'callbacks': [], 
              'post_train_args': {'user': user, 
                                  'model': model_name, 
                                  'result': output_file, 
                                  'fig_path': fig_path}}

trainer = modules.Trainer(**model_args)

In [ ]:

import warnings; 
warnings.simplefilter('ignore')

trainer.start(epoch=15)

In [ ]:

# serialize model to JSON
model_json = model.to_json()
with open("./models/metal_predict.json", "w") as json_file:
    json_file.write(model_json)
# serialize weights to HDF5
model.save_weights("./models/metal_predict.h5")
print("Saved model to disk")

from keras.models import model_from_json # load json and create model json_file = open('./models/metal_predict.json', 'r') loaded_model_json = json_file.read() json_file.close() model = model_from_json(loaded_model_json) # load weights into new model model.load_weights("./models/metal_predict.h5") print("Loaded model from disk")