Train a CNN model for filtering variants

Category Variant Filtering

---

Overview

Train a Convolutional Neural Network (CNN) for filtering variants. This tool expects requires training data generated by CNNVariantWriteTensors.

Inputs

• data-dir The training data created by CNNVariantWriteTensors.
• The --tensor-type argument determines what types of tensors the model will expect. Set it to "reference" for 1D tensors or "read_tensor" for 2D tensors.

Outputs

• output-dir The model weights file and semantic configuration json are saved here. This default to the current working directory.
• model-name The name for your model.

Usage example

Train a 1D CNN on Reference Tensors

 gatk CNNVariantTrain \
   -tensor-type reference \
   -input-tensor-dir my_tensor_folder \
   -model-name my_1d_model
 

Train a 2D CNN on Read Tensors

 gatk CNNVariantTrain \
   -input-tensor-dir my_tensor_folder \
   -tensor-type read-tensor \
   -model-name my_2d_model
 

CNNVariantTrain specific arguments

This table summarizes the command-line arguments that are specific to this tool. For more details on each argument, see the list further down below the table or click on an argument name to jump directly to that entry in the list.

Argument name(s)	Default value	Summary
Required Arguments
--input-tensor-dir		Directory of training tensors to create.
Optional Tool Arguments
--annotation-shortcut	false	Shortcut connections on the annotation layers.
--annotation-units	16	Number of units connected to the annotation input layer
--arguments_file		read one or more arguments files and add them to the command line
--conv-batch-normalize	false	Batch normalize convolution layers
--conv-dropout	0.0	Dropout rate in convolution layers
--conv-height	5	Height of convolution kernels
--conv-layers		List of number of filters to use in each convolutional layer
--conv-width	5	Width of convolution kernels
--epochs	10	Maximum number of training epochs.
--fc-batch-normalize	false	Batch normalize fully-connected layers
--fc-dropout	0.0	Dropout rate in fully-connected layers
--fc-layers		List of number of filters to use in each fully-connected layer
--gcs-max-retries  -gcs-retries	20	If the GCS bucket channel errors out, how many times it will attempt to re-initiate the connection
--gcs-project-for-requester-pays		Project to bill when accessing "requester pays" buckets. If unset, these buckets cannot be accessed. User must have storage.buckets.get permission on the bucket being accessed.
--help  -h	false	display the help message
--image-dir		Path where plots and figures are saved.
--model-name	variant_filter_model	Name of the model to be trained.
--output-dir	./	Directory where models will be saved, defaults to current working directory.
--padding	valid	Padding for convolution layers, valid or same
--spatial-dropout	false	Spatial dropout on convolution layers
--tensor-type	reference	Type of tensors to use as input reference for 1D reference tensors and read_tensor for 2D tensors.
--training-steps	10	Number of training steps per epoch.
--validation-steps	2	Number of validation steps per epoch.
--version	false	display the version number for this tool
Optional Common Arguments
--gatk-config-file		A configuration file to use with the GATK.
--QUIET	false	Whether to suppress job-summary info on System.err.
--tmp-dir		Temp directory to use.
--use-jdk-deflater  -jdk-deflater	false	Whether to use the JdkDeflater (as opposed to IntelDeflater)
--use-jdk-inflater  -jdk-inflater	false	Whether to use the JdkInflater (as opposed to IntelInflater)
--verbosity	INFO	Control verbosity of logging.
Advanced Arguments
--annotation-set	best_practices	Which set of annotations to use.
--channels-last	true	Store the channels in the last axis of tensors, tensorflow->true, theano->false
--showHidden	false	display hidden arguments

Argument details

Arguments in this list are specific to this tool. Keep in mind that other arguments are available that are shared with other tools (e.g. command-line GATK arguments); see Inherited arguments above.

---

--annotation-set / -annotation-set

Which set of annotations to use.

String  best_practices

---

--annotation-shortcut / -annotation-shortcut

Shortcut connections on the annotation layers.

boolean  false

---

--annotation-units / -annotation-units

Number of units connected to the annotation input layer

int  16  [ [ -∞  ∞ ] ]

---

--arguments_file

read one or more arguments files and add them to the command line

List[File]  []

---

--channels-last / -channels-last

Store the channels in the last axis of tensors, tensorflow->true, theano->false

boolean  true

---

--conv-batch-normalize / -conv-batch-normalize

Batch normalize convolution layers

boolean  false

---

--conv-dropout / -conv-dropout

Dropout rate in convolution layers

float  0.0  [ [ -∞  ∞ ] ]

---

--conv-height / -conv-height

Height of convolution kernels

int  5  [ [ -∞  ∞ ] ]

---

--conv-layers / -conv-layers

List of number of filters to use in each convolutional layer

List[Integer]  []

---

--conv-width / -conv-width

Width of convolution kernels

int  5  [ [ -∞  ∞ ] ]

---

--epochs / -epochs

Maximum number of training epochs.

int  10  [ [ 0  ∞ ] ]

---

--fc-batch-normalize / -fc-batch-normalize

Batch normalize fully-connected layers

boolean  false

---

--fc-dropout / -fc-dropout

Dropout rate in fully-connected layers

float  0.0  [ [ -∞  ∞ ] ]

---

--fc-layers / -fc-layers

List of number of filters to use in each fully-connected layer

List[Integer]  []

---

--gatk-config-file

A configuration file to use with the GATK.

String  null

---

--gcs-max-retries / -gcs-retries

If the GCS bucket channel errors out, how many times it will attempt to re-initiate the connection

int  20  [ [ -∞  ∞ ] ]

---

--gcs-project-for-requester-pays

Project to bill when accessing "requester pays" buckets. If unset, these buckets cannot be accessed. User must have storage.buckets.get permission on the bucket being accessed.

String  ""

---

--help / -h

display the help message

boolean  false

---

--image-dir / -image-dir

Path where plots and figures are saved.

String  null

---

--input-tensor-dir / -input-tensor-dir

Directory of training tensors to create.

R String  null

---

--model-name / -model-name

Name of the model to be trained.

String  variant_filter_model

---

--output-dir / -output-dir

Directory where models will be saved, defaults to current working directory.

String  ./

---

--padding / -padding

Padding for convolution layers, valid or same

String  valid

---

--QUIET

Whether to suppress job-summary info on System.err.

Boolean  false

---

--showHidden / -showHidden

display hidden arguments

boolean  false

---

--spatial-dropout / -spatial-dropout

Spatial dropout on convolution layers

boolean  false

---

--tensor-type / -tensor-type

Type of tensors to use as input reference for 1D reference tensors and read_tensor for 2D tensors.

The --tensor-type argument is an enumerated type (TensorType), which can have one of the following values:

reference

1 Hot encoding of a reference sequence.

read_tensor

Read tensor are 3D tensors spanning aligned reads, sites and channels. The maximum number of reads is a hyper-parameter typically set to 128. There are 15 channels in the read tensor. They correspond to the reference sequence data (4), read sequence data (4), insertions and deletions (2) read flags (4) and mapping quality (1).

TensorType  reference

---

--tmp-dir

Temp directory to use.

GATKPath  null

---

--training-steps / -training-steps

Number of training steps per epoch.

int  10  [ [ 0  ∞ ] ]

---

--use-jdk-deflater / -jdk-deflater

Whether to use the JdkDeflater (as opposed to IntelDeflater)

boolean  false

---

--use-jdk-inflater / -jdk-inflater

Whether to use the JdkInflater (as opposed to IntelInflater)

boolean  false

---

--validation-steps / -validation-steps

Number of validation steps per epoch.

int  2  [ [ 0  ∞ ] ]

---

--verbosity / -verbosity

Control verbosity of logging.

The --verbosity argument is an enumerated type (LogLevel), which can have one of the following values:

ERROR

WARNING

INFO

DEBUG

LogLevel  INFO

---

--version

display the version number for this tool

boolean  false

---

Return to top

---

GATK version 4.2.2.0-SNAPSHOT built at Thu, 19 Aug 2021 09:49:28 -0700.