Merge pull request #23 from Goosang-Yu/develop

Develop search method in SpCas9 class

README.md

@@ -101,6 +101,26 @@ df_out = spcas.predict(list_target)
 | 1      | CCTTCGTTTTTTTCCTTCTGCAGGAGGACA | CGTTTTTTTCCTTCTGCAGG | 2.253288 |
 | 2      | CTTTCAAGAACTCTTCCACCTCCATGGTGT | CAAGAACTCTTCCACCTCCA | 53.43182 |
 
+Alternatively, you can identify all possible SpCas9 target sites within an extensive gene sequence and obtain predictive scores.
+```python
+from genet.predict import SpCas9
+
+# Put the whole sequence context that you want to find Cas9 target site.
+gene = 'ttcagctctacgtctcctccgagagccgcttcaacaccctggccgagttggttcatcatcattcaacggtggccgacgggctcatcaccacgctccattatccagccccaaagcgcaacaagcccactgtctatggtgtgtcccccaactacgacaagtgggagatggaacgcacggacatcaccatgaagcacaagctgggcgggggccagtacggggaggtgtacgagggcgtgtggaagaaatacagcctgacggtggccgtgaagaccttgaaggtagg'
+
+spcas = SpCas9()
+df_out = spcas.search(gene)
+
+>>> df_out.head()
+```
+|   | Target                         | Spacer               | Strand | Start | End | SpCas9   |
+| - | ------------------------------ | -------------------- | ------ | ----- | --- | -------- |
+| 0 | CCTCCGAGAGCCGCTTCAACACCCTGGCCG | CGAGAGCCGCTTCAACACCC | +      | 15    | 45  | 67.39446 |
+| 1 | GCCGCTTCAACACCCTGGCCGAGTTGGTTC | CTTCAACACCCTGGCCGAGT | +      | 24    | 54  | 27.06508 |
+| 2 | CCGAGTTGGTTCATCATCATTCAACGGTGG | GTTGGTTCATCATCATTCAA | +      | 42    | 72  | 34.11356 |
+| 3 | AGTTGGTTCATCATCATTCAACGGTGGCCG | GGTTCATCATCATTCAACGG | +      | 45    | 75  | 76.43662 |
+| 4 | TCATCATCATTCAACGGTGGCCGACGGGCT | CATCATTCAACGGTGGCCGA | +      | 52    | 82  | 29.63767 |
+
 
 ## Tutorial 2: Predict SpCas9variants activity (by DeepSpCas9variants)
 DeepSpCas9 is a prediction model developed to evaluate to indel frequency introduced by sgRNAs at specific target sites mediated by the SpCas9 PAM variants ([Kim et al. Nat.Biotechnol. 2020](https://doi.org/10.1038/s41587-020-0537-9)). The model was developed on tensorflow (version >= 2.6). Any dependent packages will be installed along with the GenET package.

genet/models/constants.py

@@ -10,14 +10,18 @@
     'SpCas9': {
         'type': 'DeepSpCas9',
         'repo': 'Goosang-Yu/genet-models/main/genet_models',
-        'path': 'DeepSpCas9'
+        'path': 'DeepSpCas9',
+        'regex': {'+': '[ATGC]{25}GG[ATGC]{3}',
+                  '-': '[ATGC]{3}CC[ATGC]{25}',},
     },
 
     # DeepSpCas9variants
     'SpCas9-NG': {
         'type': 'DeepSpCas9variants',
         'repo': 'Goosang-Yu/genet-models/main/genet_models',
-        'path': 'DeepSpCas9variants/PAM_variant_NG'
+        'path': 'DeepSpCas9variants/PAM_variant_NG',
+        'regex': {'+': '[ATGC]{25}G[ATGC]{4}',
+                  '-': '[ATGC]{4}C[ATGC]{25}',},
     },
     'SpCas9-NRCH': {
         'type': 'DeepSpCas9variants',

genet/models/functional.py

@@ -27,13 +27,13 @@ def __init__(self, model:str, effector:str, cell_type=None):
 
         # 이 모델이 genet에서 지원하는 것인지 확인하기
         try: 
-            self.model_info = models.constants.dict_model_info[model_type]
+            self.info = models.constants.dict_model_info[model_type]
         except:
             print('[Warning] Not available model in GenET!')
             sys.exit()
 
         # model_dir: 
-        self.model_dir  = inspect.getfile(models).replace('__init__.py', '') + self.model_info['path']
+        self.model_dir  = inspect.getfile(models).replace('__init__.py', '') + self.info['path']
 
         # 만약 모델이 아직 다운로드 되지 않았다면, 다운로드 하기.
         if not os.path.exists(self.model_dir):
@@ -42,9 +42,9 @@ def __init__(self, model:str, effector:str, cell_type=None):
             dict_files = models.constants.dict_model_requests
 
             self.download_from_github(
-                repo      = self.model_info['repo'],
-                path      = self.model_info['path'],
-                files     = dict_files[self.model_info['type']],
+                repo      = self.info['repo'],
+                path      = self.info['path'],
+                files     = dict_files[self.info['type']],
                 save_dir  = self.model_dir,
                 )
 

genet/predict/DeepSpCas9.py

@@ -1,4 +1,4 @@
-import os, sys
+import os, sys, regex
 import numpy as np
 import pandas as pd
 
@@ -26,19 +26,18 @@ def __init__(self, gpu_env=0):
                             'CCTTCGTTTTTTTCCTTCTGCAGGAGGACA',
                             'CTTTCAAGAACTCTTCCACCTCCATGGTGT',
                             ]
-
-        >>> list_out = spcas9_score(list_target30)
+        >>> deepspcas9 = genet.predict.SpCas9()
         
-        >>> list_out = [2.80322408676147, 2.25273704528808, 53.4233360290527]
+        >>> spcas_score = deepspcas9(list_target30)
         '''
 
         # TensorFlow config
         self.conf = tf.compat.v1.ConfigProto()
         self.conf.gpu_options.allow_growth = True
         os.environ['CUDA_VISIBLE_DEVICES'] = '%d' % gpu_env
 
-        model_info = LoadModel('DeepSpCas9', 'SpCas9')
-        model_dir  = model_info.model_dir
+        self.model = LoadModel('DeepSpCas9', 'SpCas9')
+        model_dir  = self.model.model_dir
         best_model = 'PreTrain-Final-3-5-7-100-70-40-0.001-550-80-60'
 
         self.model_save = '%s/%s' % (model_dir, best_model)
@@ -59,12 +58,12 @@ def predict(self, list_target30: list) -> pd.DataFrame:
 
         with tf.compat.v1.Session(config=self.conf) as sess:
             sess.run(tf.compat.v1.global_variables_initializer())
-            model = DeepCas9(self.params[0], self.params[1], 80, 60, self.params[2])
+            interpreter = DeepCas9(self.params[0], self.params[1], 80, 60, self.params[2])
 
             saver = tf.compat.v1.train.Saver()
             saver.restore(sess, self.model_save)
 
-            list_score = Model_Finaltest(sess, seq_processed, model)
+            list_score = Model_Finaltest(sess, seq_processed, interpreter)
 
         df_out = pd.DataFrame()
         df_out['Target'] = list_target30
@@ -73,6 +72,50 @@ def predict(self, list_target30: list) -> pd.DataFrame:
 
         return df_out
 
+    def search(self, seq: str) -> pd.DataFrame:
+        '''주어진 sequence 내에 가능한 모든 target sequence를 찾고, 
+        그 정보와 예측 점수를 계산하는 method
+        '''
+
+        self.seq = seq.upper()
+        dict_re  = self.model.info['regex']
+
+        seq_target, seq_guide, seq_strand, pos_start, pos_end = [], [], [], [], []
+
+        for strand in ['+', '-']:
+            ptn = dict_re[strand]
+
+            for re_idx in regex.finditer(ptn, self.seq, overlapped=True):
+                if strand == '+': match = re_idx.group()
+                else            : match = reverse_complement(re_idx.group())
+
+                seq_target.append(match)
+                seq_guide.append(match[4:24])
+                seq_strand.append(strand)
+                pos_start.append(re_idx.start())
+                pos_end.append(re_idx.end())
+
+
+        seq_processed = preprocess_seq(seq_target, 30)
+
+        with tf.compat.v1.Session(config=self.conf) as sess:
+            sess.run(tf.compat.v1.global_variables_initializer())
+            interpreter = DeepCas9(self.params[0], self.params[1], 80, 60, self.params[2])
+
+            saver = tf.compat.v1.train.Saver()
+            saver.restore(sess, self.model_save)
+
+            list_score = Model_Finaltest(sess, seq_processed, interpreter)
+
+        df_out = pd.DataFrame({'Target': seq_target,
+                               'Spacer': seq_guide,
+                               'Strand': seq_strand,
+                               'Start' : pos_start,
+                               'End'   : pos_end,
+                               'SpCas9': list_score})
+
+        return df_out
+
 
 def Model_Finaltest(sess, TEST_X, model):
     test_batch = 500

genet/predict/DeepSpCas9Variants.py

@@ -1,8 +1,8 @@
 # Reference: https://blog.naver.com/PostView.naver?blogId=seodaewoo&logNo=222043145688&parentCategoryNo=&categoryNo=62&viewDate=&isShowPopularPosts=false&from=postView
-
-import tensorflow as tf
+import regex
 import numpy as np
 import pandas as pd
+import tensorflow as tf
 
 from genet.predict.PredUtils import *
 from genet.models import LoadModel
@@ -20,13 +20,11 @@ def __init__(self, effector:str):
                         'CCTTCGTTTTTTTCCTTCTGCAGGAGGACA',
                         'CTTTCAAGAACTCTTCCACCTCCATGGTGT',
                         ]
-        \n
         '''
 
-        self.effector = effector
-
-        self.model_info = LoadModel('DeepSpCas9variants', effector)
-        self.model_dir  = self.model_info.model_dir
+        self.effector  = effector
+        self.model     = LoadModel('DeepSpCas9variants', effector)
+        self.model_dir = self.model.model_dir
 
 
     def predict(self, list_target30: list) -> pd.DataFrame:
@@ -71,7 +69,51 @@ def predict(self, list_target30: list) -> pd.DataFrame:
 
         df_out[self.effector] = list_out
 
-        return df_out   
+        return df_out
+
+    def search(self, seq: str) -> pd.DataFrame:
+        '''주어진 sequence 내에 가능한 모든 target sequence를 찾고, 
+        그 정보와 예측 점수를 계산하는 method
+        '''
+
+        self.seq = seq.upper()
+        dict_re  = self.model.info['regex']
+
+        seq_target, seq_guide, seq_strand, pos_start, pos_end = [], [], [], [], []
+
+        for strand in ['+', '-']:
+            ptn = dict_re[strand]
+
+            for re_idx in regex.finditer(ptn, self.seq, overlapped=True):
+                if strand == '+': match = re_idx.group()
+                else            : match = reverse_complement(re_idx.group())
+
+                seq_target.append(match)
+                seq_guide.append(match[4:24])
+                seq_strand.append(strand)
+                pos_start.append(re_idx.start())
+                pos_end.append(re_idx.end())
+
+
+        seq_processed = preprocess_seq(seq_target, 30)
+
+        with tf.compat.v1.Session(config=self.conf) as sess:
+            sess.run(tf.compat.v1.global_variables_initializer())
+            interpreter = DeepCas9(self.params[0], self.params[1], 80, 60, self.params[2])
+
+            saver = tf.compat.v1.train.Saver()
+            saver.restore(sess, self.model_save)
+
+            list_score = Model_Finaltest(sess, seq_processed, interpreter)
+
+        df_out = pd.DataFrame({'Target': seq_target,
+                               'Spacer': seq_guide,
+                               'Strand': seq_strand,
+                               'Start' : pos_start,
+                               'End'   : pos_end,
+                               'SpCas9': list_score})
+
+        return df_out
 
 
 

genet/predict/PredUtils.py

@@ -22,4 +22,13 @@ def preprocess_seq(data, seq_length):
                 print("[Input Error] Non-ATGC character " + data[l])
                 sys.exit()
 
-    return seq_onehot
+    return seq_onehot
+
+def reverse_complement(sSeq):
+    dict_sBases = {'A': 'T', 'C': 'G', 'G': 'C', 'T': 'A', 'N': 'N', 'U': 'U', 'n': '',
+                   '.': '.', '*': '*', 'a': 't', 'c': 'g', 'g': 'c', 't': 'a'}
+    list_sSeq = list(sSeq)  # Turns the sequence in to a gigantic list
+    list_sSeq = [dict_sBases[sBase] for sBase in list_sSeq]
+    return ''.join(list_sSeq)[::-1]
+
+# def END: reverse_complement

genet/predict/functional.py

@@ -1,6 +1,7 @@
 # from genet.utils import *
 import genet
 import genet.utils
+from genet.predict.PredUtils import *
 
 import torch
 import torch.nn.functional as F
@@ -116,28 +117,6 @@ def Model_Finaltest(sess, TEST_X, model):
 
 
 
-def preprocess_seq(data, seq_length):
-
-    seq_onehot = np.zeros((len(data), 1, seq_length, 4), dtype=float)
-
-    for l in range(len(data)):
-        for i in range(seq_length):
-            try:
-                data[l][i]
-            except Exception:
-                print(data[l], i, seq_length, len(data))
-
-            if   data[l][i] in "Aa":  seq_onehot[l, 0, i, 0] = 1
-            elif data[l][i] in "Cc":  seq_onehot[l, 0, i, 1] = 1
-            elif data[l][i] in "Gg":  seq_onehot[l, 0, i, 2] = 1
-            elif data[l][i] in "Tt":  seq_onehot[l, 0, i, 3] = 1
-            elif data[l][i] in "Xx":  pass
-            elif data[l][i] in "Nn.": pass
-            else:
-                print("[Input Error] Non-ATGC character " + data[l])
-                sys.exit()
-
-    return seq_onehot
 
 def spcas9_score_tf2(list_target30:list, gpu_env=0):
     '''Tensorflow2 version function
@@ -245,15 +224,6 @@ def spcas9_score(list_target30:list, gpu_env=0):
     return list_score
 
 
-def reverse_complement(sSeq):
-    dict_sBases = {'A': 'T', 'C': 'G', 'G': 'C', 'T': 'A', 'N': 'N', 'U': 'U', 'n': '',
-                   '.': '.', '*': '*', 'a': 't', 'c': 'g', 'g': 'c', 't': 'a'}
-    list_sSeq = list(sSeq)  # Turns the sequence in to a gigantic list
-    list_sSeq = [dict_sBases[sBase] for sBase in list_sSeq]
-    return ''.join(list_sSeq)[::-1]
-
-# def END: reverse_complement
-
 def set_alt_position_window(sStrand, sAltKey, nAltIndex, nIndexStart, nIndexEnd, nAltLen):
     if sStrand == '+':