I am confusion about your deep learning architecture. about deepvariant

[lucasbrambrink]

Thank you for the question!

Take a look at how we instantiate InceptionV3 keras_modeling.py. The input_shape is inferred from the examples. InceptionV3 can handle any number of channels provided you are not using the imagenet classifier.

documentation mentions 3 channels if you are using the imagenet preset, and load the weights pre-trained on ImageNet. This is not the case if you set weights=None.

[pioneer-pi]

@lucasbrambrink Thank you for your reply! Now If I want to use the pretrained Inceptionv3 of deepvariant, where can i get the model?

[lucasbrambrink]

Sure! You can find the checkpoints for each sequencing technology at gs://deepvariant/models/DeepVariant/1.6.1/checkpoints/. For example, the model for Illumina data can be found at gs://deepvariant/models/DeepVariant/1.6.1/checkpoints/wgs/deepvariant.wgs.ckpt. These models are mounted in our Dockerfile

Take a look at our custom training case study if you want to learn more.