
# conda create -n deeplearning_py37_tf115 python=3.7
# conda activate deeplearning_py37_tf115
# pip install tensorflow-gpu==1.15
# conda install cudnn
# conda install -c conda-forge shap
# conda install matplotlib
# conda install ipykernel
# conda install numpy==1.19.5
# conda install h5py==2.10.0


print('Loading utils and functions...')

import tensorflow as tf
tf.disable_eager_execution()
tf.logging.set_verbosity(tf.logging.ERROR)
import utils
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

Loading utils and functions...


import pickle
print('Loading data...')
f = open('./data/DeepFlyBrain_data.pkl', "rb")
DeepFlyBrain_data_dict = pickle.load(f)
f.close()

Loading data...


print('Loading model...')

name = "DeepFlyBrain"
model_json_file = "./models/DeepFlyBrain/DeepFlyBrain_architecture.json"
model_hdf5_file = "./models/DeepFlyBrain/DeepFlyBrain_weights.hdf5"

model_dict = {}
model_dict[name] = utils.load_model(model_json_file, model_hdf5_file)

Loading model...


number_of_random_regions = 100
seq_len = 500


motif_embedding_dict = {}
motif_embedding_dict["Random"] = {}
motif_embedding_dict["Random"]["regions"] = utils.random_sequence(seq_len, number_of_random_regions)


number_of_random_regions = 100
seq_to_shuffle = "AGGGTCGGGCGTGTCCGCCTGTTGGAATATGCTTCTCAGGTCTTCTGGGAAACAGATGTTTTGTGGAAGTGGAAGATTTTGGAAGTAGTGCCTTATCATGTGAAACCACAGGGCAGCTGATCTCTTCAGGCTTTCTTGATGTGAATGACAGCTTTGTTTCATCCACTTTGGTGGGTAAAAGAAGGCAAATTCCCCTGTGGTACTTTTGGTGCCAGGTTTAGCCATATGACGAAGCTTTACATAAAACAGTACAAGTATCTCCATTGTCCTTTATGATCCTCCATGAGTGTTTTCACTTAGTCTGATGAAGGGTTCACTCCAGTCTTTTCGGATGATAAAATGCTTCGGCTGTCAGTCTAATAAGGGATTCCCTGAGGAGTTTGGAGGCTGTAAGAGCACCCCCCGTCTCAATGCCAGTGCTTCTTATCTCAGCCTCTCCTGCACTCCTTTACCCCCGTCTCGATGCCAGTGCTTCCTATCTCAGCCTCTCCTGCACTCCT"


motif_embedding_dict = {}
motif_embedding_dict["Random"] = {}
motif_embedding_dict["Random"]["regions"] = utils.random_sequence_by_shuffling(seq_to_shuffle, number_of_random_regions)


path_to_use_GC_content = "./data/KC_regions.fa"
number_of_random_regions = 100
seq_len = 500


motif_embedding_dict = {}
motif_embedding_dict["Random"] = {}
motif_embedding_dict["Random"]["regions"] = utils.random_sequence_gc_adjusted(seq_len, number_of_random_regions, path_to_use_GC_content)


patterns_dict = {"Ey": "TGCTCACTCAAGCGTAA",
                 "Mef2": "CTATTTATAG",
                 "Onecut": "ATCGAT",
                 "Sr": "CCACCC"}

for factor in patterns_dict:
    patterns_dict[factor] = utils.one_hot_encode_along_row_axis(patterns_dict[factor])


path_to_save = "./output/testdata_motif_implant_100seqs.pkl"


print("Implanting Ey motif to random sequences")
motif_embedding_dict["Ey"] = utils.add_pattern_to_best_location(pattern = patterns_dict["Ey"], 
                                                                regions = motif_embedding_dict["Random"]["regions"], 
                                                                model = model_dict["DeepFlyBrain"], 
                                                                class_no = 35)

print("Implanting Mef2 motif to Ey implanted sequences")
motif_embedding_dict["Ey+Mef2"] = utils.add_pattern_to_best_location(pattern = patterns_dict["Mef2"], 
                                                                     regions = motif_embedding_dict["Ey"]["regions"], 
                                                                     model = model_dict["DeepFlyBrain"], 
                                                                     class_no = 35)

print("Scoring final sequences")
for key in motif_embedding_dict:
    motif_embedding_dict[key]["prediction"] = model_dict["DeepFlyBrain"].predict(motif_embedding_dict[key]["regions"])
    
import pickle
f = open(path_to_save, "wb")
pickle.dump(motif_embedding_dict,f)
f.close()

Implanting Ey motif to random sequences
Sequence index: 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,
Implanting Mef2 motif to Ey implanted sequences
Sequence index: 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,
Scoring final sequences


path_to_file = "./output/testdata_motif_implant_100seqs.pkl"


import pickle
f = open(path_to_file, "rb")
motif_embedding_dict = pickle.load(f)
f.close()


class_no = 35
_ = plt.boxplot([motif_embedding_dict[key]["prediction"][:,class_no-1] for key in motif_embedding_dict],notch=True,showfliers=False,whis=[5,95])
_ = plt.xticks(range(1,len(motif_embedding_dict.keys())+1),list(motif_embedding_dict.keys()))
plt.ylabel("Prediction score")

Text(0, 0.5, 'Prediction score')


_ = plt.hist(motif_embedding_dict["Ey"]["locations"],bins=500,range=(0,500))
_ = plt.ylabel("Number of sequences")
_ = plt.xlabel("Position (bp)")
_ = plt.title("Location of Ey")


_ = plt.hist(motif_embedding_dict["Ey+Mef2"]["locations"]-motif_embedding_dict["Ey"]["locations"],bins=401,range=(-200.5,200.5))
_ = plt.ylabel("Number of sequences")
_ = plt.xlabel("Position (bp)")
_ = plt.title("Location of Mef2 relative to Ey")


index = 19
class_no =  35
onehot_ = np.copy(motif_embedding_dict["Ey+Mef2"]["regions"][index:index+1])


# Initialize the explainer for the selected class
import shap
tf.disable_eager_execution()
rn=np.random.choice(DeepFlyBrain_data_dict["nonAugmented_data"].shape[0], 250, replace=False)
name = "DeepFlyBrain"
explainer_dict = {} 
explainer_dict[name+"_c"+str(class_no)] = shap.DeepExplainer((model_dict[name].inputs,
                                                              model_dict[name].layers[-1].output[:,class_no-1]),
                                                             DeepFlyBrain_data_dict["nonAugmented_data"][rn])


# Plot contribution scores and saturation mutagenesis
ntrack = 2
fig = plt.figure(figsize=(80,ntrack*5))
ax1 = utils.plot_deepexplainer_givenax(explainer=explainer_dict["DeepFlyBrain_c35"], fig=fig, ntrack=ntrack, track_no=1, seq_onehot=onehot_)
_ = utils.plot_mutagenesis_givenax(model=model_dict["DeepFlyBrain"], fig=fig, ntrack=ntrack, track_no=2, seq_onehot=onehot_, class_no = class_no)


index = 19
class_no = 35

for i,step in enumerate(motif_embedding_dict):
    plt.bar(i,motif_embedding_dict[step]["prediction"][index,class_no-1],label=step,color="gray")
_ = plt.xticks(range(len(motif_embedding_dict.keys())),list(motif_embedding_dict.keys()))
_ = plt.ylabel("Prediction score")


index = 19
onehot_ = np.copy(motif_embedding_dict["Ey+Mef2"]["regions"][index:index+1])

print("Generated sequence:",end=" ")
for nuc in onehot_[0]:
    if nuc[0]==1:
        print("A",end="")
    if nuc[1]==1:
        print("C",end="")
    if nuc[2]==1:
        print("G",end="")
    if nuc[3]==1:
        print("T",end="")
print("")

Generated sequence: TGACTTATGGGTATCGAGTAAAATAGGCAGAATCTGATGGTGGCTGTATAATATAATATATATCCCAGTATCCCCTGAATTTGTTACTGTTCTAGATTTTAAATCAAAAGCCCCGTCTTTGGCCGGCATATTAAGCTACCAACCTCTTATGGAAGGTTCAGGCCATCCGAGTGGCGATATACCAACTGGGGCTCGGCTGGGTATAATAAGTCGTTTGTCAGTGAGACCGGAACCCGGAGCACTGGACATCTGTACTCGTGTGCGTGGTACCGCTATTTATAGGCCGCATCGTTAATCCGTGCTCACTCAAGCGTAAAATAGACAGAATGGAGAGCTAAGAGGGTAGAACGAGGTTGGCGGTCGTTCTTAAATTATAGACCCATAGTTTGTTAGCCCCTGCGATACACCTCTCTTTCAGGTCGAGTATAGCATCTAGAGCACGTGGGGGTGTCCCTAGTAGTTTAAGTGGGGTACGATTGCCTTTTAATTAGCTTGTCAGT

The following script was run using a conda environment. Below you can find how to create the same environment¶

Load utils and necessary functions¶

Load the data¶

(to be used to initialize DeepExplainer)¶

Load the model¶

Create random sequences¶

(or) Create random sequences by shuffling a given sequence.¶

(Length of the given sequence should be the same size you used for the model, which is 500bp in this case)¶

(or) Create random sequences that follow the GC-content of the given fasta sequences¶

Define the patterns to be implanted¶

Here you generate synthetic sequence by in silico evolution.¶

pattern: selected motif to be implanted¶

region: initial regions to implant the selected pattern on¶

model: selected deep learning model¶

class_no: chosen class to select the best position to implant motif¶

As an output it results a dictionary; regions contains the sequences after implantation to the best location and locations contains the implanted position as the position of first nucleotide of the implanted motif¶

Load the saved file¶

Plot the prediction distributions¶

Plot the location distribution of an implanted motif¶

Plot the location distribution of an implanted motif relative to another implanted motif¶

Plot deepexplainer and saturation mutagenesis of a selected sequence¶

Plot prediction of a region at different steps¶

Convert created sequence to letters¶