DAK: Explaining the genetic causality for complex diseases by deep association kernel learning

Causal loci contribute to complex diseases in various manners. The comprehensive identification of suspicious genes requires a general genome-wide association study (GWAS) model that can work with different types of genetic effects. Here, we introduce a deep association kernel learning (DAK) model to enable automatic causal genotype encoding for pathway-level GWAS. Therefore, DAK is able to detect common and rare variants with complicated genetic effects that existing approaches fail. DAK is published as a cover paper in Patterns.

Citation: Bao, Feng, et al. "Explaining the Genetic Causality for Complex Phenotype via Deep Association Kernel Learning." Patterns 1.6 (2020): 100057. Link to the paper

Package requirements

DAK requires the following packages for installation:

• Python >= 3.6
• TensorFlow-GPU >= 1.4.1
• (TensorFlow >= 1.4.1 if only use CPU)
• Numpy >= 1.13.0
• Scipy >= 1.0.0
• Pandas >= 0.22.0
• os
• time

Usage

Installation

terminal pip install dak

Data preparation

1. Genotype data

   SNP loci in the same set are stored in one document in the n * m format, where n is the number of samples and m is the number of SNPs in the set. Each locus is in the additive genetic coding format and the DAK.one_hot_convert(geno) function will automatically transform the sequence to one-hot coding data.

   A demonstration genotype format file can be found in ./demo_data/pathway_*.raw_geno.txt

2. Phenotype data

   Phenotype information is in the n*1 vector format with each row representing the disease status of one sample. 1: disease; 0: control.

   Refer to ./demo_data/pheno.txt for example.

3. Cofounding data

   Cofounding of samples is in n*k matrix where k is the number of PCs/covariants.

4. Dividing to batches

   Dividing all data into batch files for training and inference. Users can refer to Step 5 in Demonstration for detailed implementation.

DAK parameters

The parameters of DAK function is listed as follows:

python dak = DAK(sess, # tensorflow session that conducts learning task batch_path_prefix=batch_path_prefix, # file path of genotype data in batches label_path_prefix=label_path_prefix, # file path of label data in batches cov_path_prefix=cov_path_prefix, # file path of covariant data in p_val_path=p_val_path, # file path of p-values by DAK batch_num=batch_num, # the batch number of the data batch_size=batch_size, # sample number in each batch file pathway_num=pathway_num, # number of gene sets max_path_len=max_path_len, # the maximal SNP number among all gene sets );

Demonstration

1. Define the file path of phenotype. ```python

   set the path of label data and covariants (optinal)

   labelpath = './demodata/pheno.txt'

   covpath = '../application/LCpathway/LCpathwaycov.txt'

   ```

2. Define paths of outputs: p-value, genotype batches, label batches, one-hot coded genotype (for internal usage of DAK). ```python

   set the aim paths of result and

   resultpath = './demodata/p.txt' pathwaynpypath = './demodata/pathwayonehot' batchnpypath = './demodata/batch' batchlabelpath = './demodata/label'

   batchcovpath = './demo_data/cov' # (optinal)

   ```

3. Set the details of analyzed data python pathway_num = 10 # number of gene sets indiv_num = 1000 #number of samples in total batch_size = 50 # number of samples in each batch file max_path_len = 20000 # maximal SNP numbers in all gene sets

4. One-hot coding for genotype ```python

   convert raw format SNP into one-hot coding

   rawpath = '../application/LCpathway' for pathiter in range(pathwaynum): geno = pd.readcsv('./demodata/pathway' + str(pathiter) + '.rawgeno.txt', sep='\t', header=None, indexcol=None) geno = geno.values geneonehot = DAK.onehotconvert(geno) np.save(pathwaynpypath + '/pathway' + str(pathiter) + '.npy', geneonehot) print('One hot conversion for pathway ' + str(path_iter)) ```

5. Divide data into batches ```python

   convert pathway to training batches

   batchindex = range(0, indivnum, batchsize) label = pd.readcsv(labelpath, sep='\t', header=0, indexcol=None) label = np.squeeze(label.values)

cov = pd.readcsv(covpath, sep='\t', header=0, index_col=None)

cov = cov.values.astype(np.float)

cov = stats.zscore(cov, axis=0)

divided to batches

for i in range(len(batchindex) - 1): batchseq = np.zeros( [pathwaynum, batchsize, maxpathlen, 3], dtype=np.int8) for pathiter in range(pathwaynum): pathdatabuf = np.load( pathwaynpypath + '/pathway' + str(pathiter) + '.npy') # [N,len,3] pathdatabufselect = pathdatabuf[batchindex[i]:batchindex[i + 1], :, :] batchseq[pathiter, :, :pathdatabufselect.shape[1], :] = pathdatabuf_select

batch_seq = batch_seq.astype(np.int8)
np.save(batch_npy_path + '/batch_' + str(i) + '.npy', batch_seq)

batch_label = label[batch_index[i]:batch_index[i + 1]]
np.save(batch_label_path + '/batch_' + str(i) + '.npy', batch_label)

# batch_cov = cov[batch_index[i]:batch_index[i + 1], :]
# np.save(batch_cov_path + '/batch_' + str(i) + '.npy', batch_cov)

print('make batch %d' % i)

```

1. Model training and significance test python # training DAK and test pathway DAK.train(batch_npy_path, batch_label_path, None, result_path, batch_num=len(batch_index) - 1, batch_size=batch_size, pathway_num=pathway_num, max_path_len=max_path_len) Results were store in path specified in result_path.

Example

Code

See demo.py.

Data

In ./demo_data/:

• Genotype: ./demo_data/pathway_*.raw_geno.txt
• Phenoytpe: ./demo_data/pheno.txt

Copyright

Software provided as is under MIT License.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.