sgrna_modeler package

Submodules

sgrna_modeler.datasets module

Training and testing datasets for modeling guide activity.

Includes class for generating new dataset objects, so new data can be easily modeled and tested

Example:

>>> import pandas as pd
>>> import sgrna_modeler.enzyme as en
>>> new_data = pd.read_csv('new dataset')
>>> new_dataset = Activity_Data(new_data, en.cas9, '30mer', 'activity', 'new data')

class sgrna_modeler.datasets.ActivityData(data, enzyme, kmer_column, activity_column, name, group_column='')

Bases: object

Store information about activity data

Parameters:

• data (pandas dataframe) – data to model
• enzyme (dict) – cas9 or cas12a
• kmer_column (str) – sequences to model
• name (str) – name of the dataset
• group_column – column to include in prediction output

:type group_column:str

get_xy()

Gets modeling matrix (x) and output matrix (y)

:return two series, x and y :rtype pandas series

sgrna_modeler.datasets.curr_path()

sgrna_modeler.datasets.load_doench_2016()

Data from:

Doench, John G., et al. “Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9.” Nature biotechnology 34.2 (2016): 184.

Example:

>>> import sgrna_modeler.datasets as da
>>> doench = da.load_doench_2016()
>>> doench.data
      Unnamed: 0                           30mer  ...       drug  predictions
0              0  CAGAAAAAAAAACACTGCAACAAGAGGGTA  ...     nodrug     0.544412
1              1  TTTTAAAAAACCTACCGTAAACTCGGGTCA  ...    PLX_2uM     0.617512
2              2  TCAGAAAAAGCAGCGTCAGTGGATTGGCCC  ...     nodrug     0.476232
3              3  AATAAAAAATAGGATTCCCAGCTTTGGAAG  ...    PLX_2uM     0.459882
4              4  GATGAAAAATATGTAAACAGCATTTGGGAC  ...    PLX_2uM     0.290841
          ...                             ...  ...        ...          ...
5305        5305  GCACTTTGGTGTGGCTGACTGAGTGGGCCA  ...    PLX_2uM     0.586758
5306        5306  TTCTTTTGTAAGAACCCGCTGTGTTGGTTT  ...    PLX_2uM     0.492066
5307        5307  GCCCTTTGTCATCGTAGGAAGATATGGCTG  ...  AZD_200nM     0.479728
5308        5308  CAAATTTGTTCTTTAAATGGCTACAGGAGG  ...  AZD_200nM     0.478952
5309        5309  CAAATTTGTTCTTTAAATGGCTACAGGAGG  ...    PLX_2uM     0.478952
[5310 rows x 9 columns]

sgrna_modeler.datasets.load_kim_2018_test()

Indel frequencies from:

Kim, Hui Kwon, et al. “Deep learning improves prediction of CRISPR–Cpf1 guide RNA activity.” Nature biotechnology 36.3 (2018): 239.

Example:

>>> import sgrna_modeler.datasets as da
>>> kim_2018_test = da.load_kim_2018_test()
>>> kim_2018_test.data
         50 bp synthetic target and target context sequence  ... Indel frequency
0     GCAATTTGGTTTTAAAACAGAATATACAGTCTAAAAAACCAGCTTG...  ...       71.580711
1     CTGATGGCCATTTAAACAACTCTTTGAGCTCTCCAGTTCAAGCTTG...  ...       19.672949
2     TTTAGATGATTTTAAACCAGCATCTATAGACACTTCCTGTAGCTTG...  ...       75.641026
3     ACATTTGGACTTTAAACCCAAACTACTTGTCCAACGGTACAGCTTG...  ...       46.920217
4     CTCTACCAGGTTTAAACGCTTCCACACTTGTGTCAGTAATAGCTTG...  ...       54.981550
                                                 ...  ...             ...
2958  AGTTTGGAATTTTTTTTACACTGATCCTCAGCACATCTCAAGCTTG...  ...       -0.378500
2959  CAGGCTTTCTTTTTTTTCCTTTCCTAGTTGGTTCATTCCCAGCTTG...  ...        0.189438
2960  AACAGTGGCTTTTTTTTGCTGCTAGCACATATGTATGGGTAGCTTG...  ...       -2.857143
2961  CAGCCTCATGTTTTTTTGGGAACCAATCGATAATCACATTAGCTTG...  ...       11.275673
2962  TTGGATTGTGTTTTTTTTTAGCACCTTATTTTCCTTGAAGAGCTTG...  ...       -1.675978
[2963 rows x 10 columns]

sgrna_modeler.datasets.load_kim_2018_train()

Indel frequencies from:

Kim, Hui Kwon, et al. “Deep learning improves prediction of CRISPR–Cpf1 guide RNA activity.” Nature biotechnology 36.3 (2018): 239.

Example:

>>> import sgrna_modeler.datasets as da
>>> kim_2018_train = da.load_kim_2018_train()
>>> kim_2018_train.data
          50 bp synthetic target and target context sequence 10 bp + PAM + 23 bp protospacer + 17 bp)  ... Indel frequency
0      TGCGCGAGCGTTTAAAAAACATCGAACGCATCTGCTGCCTAGCTTG...                                             ...       14.711302
1      CTAAAGAAACTTTAAAAATCTTTTCTGCCAGATCTCCAGAAGCTTG...                                             ...        0.238095
2      TTGCCATTGTTTTAAAACAGGTTCTGTACTTGATCTCTCCAGCTTG...                                             ...       88.079746
3      TTGCACATATTTTAAAACTGAGTTCAAAGACCACTCTTCCAGCTTG...                                             ...       75.392670
4      TAGACTAATGTTTAAAAGCAAGTGCAAGTCTTTGGAATCTAGCTTG...                                             ...       63.320080
                                                  ...                                             ...             ...
14995  TCCATCTTCATTTTTTTTGTAGAGTAGGGCTTTATTTCCAAGCTTG...                                             ...       -0.467290
14996  CCTTCTCTCCTTTTTTTTTCAAGATCTGATTCTTCTTGCAAGCTTG...                                             ...        0.000000
14997  CCAGGACTTGTTTTTTTTTCAATCTGTTCATCTTGGACCAAGCTTG...                                             ...        0.239006
14998  ACCATCATAATTTTTTTTTGCAACATAGCCATTTCTTTTTAGCTTG...                                             ...       -0.272826
14999  GAGCGCTTCTTTTTTTTTTTCGGGGTCTCGTTGCTGGGCGAGCTTG...                                             ...       -2.766164
[15000 rows x 10 columns]

sgrna_modeler.datasets.load_kim_2019_test()

Indel frequencies from:

Kim, Hui Kwon, et al. “SpCas9 activity prediction by DeepSpCas9, a deep learning–based model with high generalization performance.” Science advances 5.11 (2019): eaax9249.

Example:

>>> import sgrna_modeler.datasets as da
>>> kim_2019_test = da.load_kim_2019_test()
>>> kim_2019_test.data
    Target context sequence (4+20+3+3)  ...  Background subtracted indel frequencies (average, %)
0       AAAACTGTGAGTGTGGGACCTGCTGGGGGC  ...                                          44.125755
1       AAACACAACCAATCCGAGGCCTTCTGGGTC  ...                                          12.163189
2       AAACTGTGAGTGTGGGACCTGCTGGGGGCT  ...                                          68.901263
3       AAACTTGAGAGCTTTCATAAAGCTTGGCAA  ...                                          13.135690
4       AAAGAAGCGGACTTTAAAGTTCGAGGGAGA  ...                                          48.355156
..                                 ...  ...                                                ...
537     TTTGCAGCGCGTTGACTTATTCATGGGTCA  ...                                          36.249050
538     TTTGCTAGGAATATTGAAGGGGGCAGGGGA  ...                                          38.622947
539     TTTGTGGTGGTTGCTATGGTAATCCGGCAC  ...                                          12.246218
540     TTTTTACAATTCTGTGAGTTAGAGTGGGCA  ...                                           0.385915
541     TTTTTGAGGTGCACTAATAGAGGGTGGAGT  ...                                          41.100730
[542 rows x 5 columns]

sgrna_modeler.datasets.load_kim_2019_train()

Indel frequencies from:

Kim, Hui Kwon, et al. “SpCas9 activity prediction by DeepSpCas9, a deep learning–based model with high generalization performance.” Science advances 5.11 (2019): eaax9249.

Example:

>>> import sgrna_modeler.datasets as da
>>> kim_2019_train = da.load_kim_2019_train()
>>> kim_2019_train.data
                       Barcode  ... Background subtracted indel (%)
0      TTTGACACACACGCACTAG  ...                       24.287805
1      TTTGACACACACTCGTATG  ...                       69.500438
2      TTTGACACACACTCTCGTC  ...                       25.994760
3      TTTGACACACACTCTGCTG  ...                       57.964590
4      TTTGACACACACTGCATAT  ...                       39.355020
                    ...  ...                             ...
12827  TTTGTGTGTCTCGTATCAC  ...                       40.853256
12828  TTTGTGTGTCTCTACACGC  ...                       11.480880
12829  TTTGTGTGTCTCTCACGTA  ...                       63.861469
12830  TTTGTGTGTCTCTCTAGTC  ...                       51.650932
12831  TTTGTGTGTCTCTCTCAGA  ...                       40.019124
[12832 rows x 9 columns]

sgrna_modeler.datasets.load_meyers_2017_test()

Essential genes from GeckoV2 achilles screens:

Meyers, Robin M., et al. “Computational correction of copy number effect improves specificity of CRISPR–Cas9 essentiality screens in cancer cells.” Nature genetics 49.12 (2017): 1779-1784.

Mean activity is averaged accross screens after Z-scoring by non-essentials

Example:

>>> import sgrna_modeler.datasets as da
>>> meyers_2017_test = da.load_meyers_2017_test()
>>> meyers_2017_test.data
            Species   Build  Chromosome Number  ... Percent Protein Notes mean_activity
0     human  GRCh38                1.0  ...           22.12   NaN      3.325952
1     human  GRCh38                1.0  ...            2.30   NaN      2.645421
2     human  GRCh38                1.0  ...           56.87   NaN      2.040191
3     human  GRCh38                1.0  ...           40.38   NaN      3.356250
4     human  GRCh38                1.0  ...           40.11   NaN      1.602670
..      ...     ...                ...  ...             ...   ...           ...
667   human  GRCh38                NaN  ...            8.56   NaN      1.240547
668   human  GRCh38                NaN  ...            8.95   NaN      1.078080
669   human  GRCh38                NaN  ...           30.93   NaN     -0.364154
670   human  GRCh38                NaN  ...           34.24   NaN      2.605412
671   human  GRCh38                NaN  ...           41.25   NaN      2.620977
[672 rows x 20 columns]

sgrna_modeler.datasets.load_meyers_2017_train()

Essential genes from GeckoV2 achilles screens:

Meyers, Robin M., et al. “Computational correction of copy number effect improves specificity of CRISPR–Cas9 essentiality screens in cancer cells.” Nature genetics 49.12 (2017): 1779-1784.

Mean activity is averaged accross screens after Z-scoring by non-essentials

Example:

>>> import sgrna_modeler.datasets as da
>>> meyers_2017_train = da.load_meyers_2017_train()
>>> meyers_2017_train.data
         Species   Build  Chromosome Number  ... Percent Protein Notes mean_activity
0      human  GRCh38                1.0  ...           24.87   NaN     -0.230160
1      human  GRCh38                1.0  ...           23.26   NaN      3.045755
2      human  GRCh38                1.0  ...           18.60   NaN      1.307097
3      human  GRCh38                1.0  ...           18.07   NaN     -1.307698
4      human  GRCh38                1.0  ...           13.95   NaN      1.278670
      ...     ...                ...  ...             ...   ...           ...
7897   human  GRCh38                NaN  ...           11.78   NaN      1.959897
7898   human  GRCh38                NaN  ...           13.14   NaN     -0.429659
7899   human  GRCh38                NaN  ...           16.01   NaN      1.187820
7900   human  GRCh38                NaN  ...           19.03   NaN      1.573194
7901   human  GRCh38                NaN  ...           39.62   NaN      2.044455
[7902 rows x 20 columns]

sgrna_modeler.enzymes module

sgrna_modeler.features module

sgrna_modeler.features.encode_seqs(seqs)

sgrna_modeler.features.featurize_guides(kmers, features=None, guide_start=9, guide_length=20)

Take guides and encodes for modeling :param kmers: vector of context sequences :param features: boolean dictionary of which feature types to inlcude:

[‘Pos. Ind. 1mer’, ‘Pos. Ind. 2mer’, ‘Pos. Ind. 3mer’, ‘Pos. Ind. Zipper’,’Pos. Dep. 1mer’, ‘Pos. Dep. 2mer’, ‘Pos. Dep. 3mer’, ‘Pos. Dep. Zipper’, ‘Pos. Ind. Rep.’, ‘GC content’, ‘Tm’, ‘Physio’, ‘Double Zipper’]

Parameters:

• pam_start – int
• pam_end – int
• guide_start – int
• guide_end – int

Returns:

featurized matrix

sgrna_modeler.features.get_context_order(k)

Parameters:k – length of kmer Returns:list of characters of each nt position

sgrna_modeler.features.get_double_zipper(dict, context_sequence, nts, context_order)

sgrna_modeler.features.get_frac_g_or_c(dict, guide_sequence)

Get gc content :param context: sequence :param len: length of guide :param guide_start: position that guide starts :return: curr_dict with gc content

sgrna_modeler.features.get_guide_sequence(context, guide_start, guide_length)

sgrna_modeler.features.get_one_nt_counts(dict, guide, nts)

sgrna_modeler.features.get_one_nt_pos(dict, context_sequence, nts, context_order)

sgrna_modeler.features.get_physiochemical(curr_dict, guide, nts, physiochemical_data)

sgrna_modeler.features.get_rep_counts(dict, context, nts, length)

sgrna_modeler.features.get_thermo(dict, guide_sequence, context_sequence)

sgrna_modeler.features.get_three_nt_counts(dict, guide, nts)

sgrna_modeler.features.get_three_nt_pos(dict, context_sequence, nts, context_order)

sgrna_modeler.features.get_two_nt_counts(dict, guide, nts)

sgrna_modeler.features.get_two_nt_pos(dict, context_sequence, nts, context_order)

sgrna_modeler.features.get_zipper_counts(dict, guide, nts)

sgrna_modeler.features.get_zipper_pos(dict, context_sequence, nts, context_order)

sgrna_modeler.models module

class sgrna_modeler.models.KerasSgrnaModel(random_state=7, val_frac=0.1, base_arc=None)

Bases: object

This class is for creating, training, and predicting guide activity with a Keras model

Parameters:

• random_state – set random state in train/test split for reproducibility
• val_frac (float) – amount of data to use for early stopping
• base_arc (function, which takes an input shape and returns a keras model) – base architecture to build neural network, defaults to build_kim2018

Example:

>>> from sgrna_modeler import datasets as da
>>> from sgrna_modeler import models as sg
>>> train_data = da.load_kim_2018_train()
>>> train_model = sg.KerasSgrnaModel()
>>> train_model.fit(train_data)
>>> test_data = da.load_kim_2018_test()
>>> test_predictions = train_model.predict(test_data)

fit(train_dataset)

Fit a model to the training data

Parameters:train_dataset (sgrna_modeler.datasets.ActivityData) – training data Returns:self

load_weights(weights, enzyme, name)

Load previously trained weights

Parameters:

• enzyme – cas9 or cas12a
• weights (str) – filepath to weights
• name – name of the model

:type name:str

predict(test_dataset)

Predict activity of test data

Parameters:test_dataset (sgrna_modeler.datasets.ActivityData) – testing data Returns:dataframe of predictions and other meta information Return type:pandas dataframe

predict_seqs(seqs)

Predict from sequences

Parameters:seqs – sequences to predict Returns:numeric vector of predcitions

class sgrna_modeler.models.SklearnSgrnaModel(random_state=7, val_frac=0.1, model=None, features=None)

Bases: object

scikit-learn gradient boosting for modeling sgRNA activity

Parameters:

• random_state (int) – set random state in train/test split for reproducibility
• val_frac (float) – amount of data to use for early stopping
• model (sklearn GradientBoostingRegressor) – base model
• features (list) – features to model

Example:

>>> from sgrna_modeler import datasets as da
>>> from sgrna_modeler import models as sg
>>> train_model = sg.SklearnSgrnaModel()
>>> rs2_data = da.load_doench_2016()
>>> train_model.fit(rs2_data)

fit(train_dataset)

Fit a model to the training data

Parameters:train_dataset (sgrna_modeler.datasets.ActivityData) – training data Returns:self

load_model(model, enzyme, name)

Load previously trained model

Parameters:

• enzyme – cas9 or cas12a
• model (str (*.joblib)) – filepath to trained model
• name – name of the model

:type name:str

predict(test_dataset)

Predict activity of test data

Parameters:test_dataset (sgrna_modeler.datasets.ActivityData) – testing data Returns:dataframe of predictions and other meta information Return type:pandas dataframe

predict_seqs(seqs)

Predict from sequences

Parameters:seqs – sequences to predict Returns:numeric vector of predcitions

sgrna_modeler.models.build_kim2018(input_shape=(34, 4))

Build a convolutional neural network

From: Kim, Hui Kwon, et al. “Deep learning improves prediction of CRISPR–Cpf1 guide RNA activity.” Nature biotechnology 36.3 (2018): 239.

Parameters:input_shape (tuple) – guide length by nts (4) Returns:CNN architecture Return type:keras Model object

sgrna_modeler.models.curr_path()

sgrna_modeler.models.get_deepcpf1_weights()

sgrna_modeler.models.get_enpam_gb()

sgrna_modeler.mutagenesis module

sgrna_modeler.mutagenesis.generate_variants(seq)

sgrna_modeler.mutagenesis.mutagenize_model(model, resamples)

sgrna_modeler.mutagenesis.mutagenize_seq(guide, model)

Module contents

Top-level package for sgRNA Modeler.