We've moved!
This site is now read-only. You can find our new documentation site and support forum for posting questions here.
Be sure to read our welcome blog!

gatk CNNScoreVariants cannot reset read-limit


I tried CNNScoreVariants and it went well when I used all default settings. But when I reset --read-limit to 2048 instead of default value 128, the process failed. The following is the end of the journal file:


nkm received

Traceback (most recent call last):
File "", line 1, in
File "/home/ying/anaconda2/envs/gatk/lib/python3.6/site-packages/vqsr_cnn/vqsr_cnn/inference.py", line 130, in score_and_write_bat
[np.array(read_batch), np.array(annotation_batch)], batch_size=python_batch_size)
File "/home/ying/anaconda2/envs/gatk/lib/python3.6/site-packages/keras/engine/training.py", line 1152, in predict
x, _, _ = self._standardize_user_data(x)
File "/home/ying/anaconda2/envs/gatk/lib/python3.6/site-packages/keras/engine/training.py", line 754, in _standardize_user_data
File "/home/ying/anaconda2/envs/gatk/lib/python3.6/site-packages/keras/engine/training_utils.py", line 136, in standardize_input_d
ValueError: Error when checking input: expected read_tensor to have shape (128, 128, 15) but got array with shape (2048, 128, 15)
Received from stderr: [Traceback (most recent call last):
File "", line 1, in

I tried to reset the read-limit because my data are amplicon based sequencing results and have extremely high depth. I checked all parameters, and did not see any other parameter that could relate to read depth.

Here is a sample of my job script:

gatk --java-options "-Djava.io.tmpdir=/mnt/scratch -XX:GCTimeLimit=50 -XX:GCHeapFreeLimit=10 -Xms4G -Xmx60G" CNNScoreVariants
-I /cnn/IonXpress_056.9.hc.bam -R /refGenome/gatk/hg19bundle/ucsc.hg19.fasta -V /cnn/IonXpr
ess_056.9.hc.raw.vcf -O /cnn/IonXpress_056.9.hc.cnn.2048.vcf -L /refGenome/PanelTargets/IAD174034_166_Desi
gned.bed --interval-padding 100 --tensor-type read_tensor --inference-batch-size 8 --transfer-batch-size 32 --read-limit 2048 --enable-journal

Any suggestion?




Sign In or Register to comment.