Deep learning-based techniques for training deep convolutional neural networks
The technology disclosed relates to constructing a convolutional neural network-based classifier for variant classification. In particular, it relates to training a convolutional neural network-based classifier on training data using a backpropagation-based gradient update technique that progressively match outputs of the convolutional network network-based classifier with corresponding ground truth labels. The convolutional neural network-based classifier comprises groups of residual blocks, each group of residual blocks is parameterized by a number of convolution filters in the residual blocks, a convolution window size of the residual blocks, and an atrous convolution rate of the residual blocks, the size of convolution window varies between groups of residual blocks, the atrous convolution rate varies between groups of residual blocks. The training data includes benign training examples and pathogenic training examples of translated sequence pairs generated from benign variants and pathogenic variants.
1. A computer-implemented method of constructing a convolutional neural network-based classifier for variant classification, the method including:
training a convolutional neural network-based classifier, which runs on numerous processors coupled to memory, on training data using a backpropagation-based gradient update technique that progressively matches outputs of the convolutional neural network-based classifier with corresponding ground truth labels;
wherein the convolutional neural network-based classifier comprises groups of residual blocks;
wherein each group of residual blocks is parameterized by a number of convolution filters in the residual blocks, a convolution window size of the residual blocks, and an atrous convolution rate of the residual blocks;
wherein the convolution window size varies between the groups of residual blocks;
wherein the atrous convolution rate varies between the groups of residual blocks;
wherein the training data includes pairs of translated sequences generated from benign variants and pathogenic variants and used as benign training examples and as pathogenic training examples; and
wherein the benign variants include common human missense variants and non-human primate missense variants occurring on alternative non-human primate base triplet sequences that share matching reference base triplet sequences with humans.
2. The computer-implemented method of claim 1 , wherein each residual block comprises at least one batch normalization layer, at least one rectified linear unit (ReLU) layer, at least one atrous convolution layer, and at least one residual connection.
3. The computer-implemented method of claim 1 , wherein the convolutional neural network-based classifier is parameterized by a number of residual blocks, a number of skip connections, and a number of residual connections.
4. The computer-implemented method of claim 1 , further including one-hot encoding the training examples and providing one-hot encodings as input to the convolutional neural network-based classifier.
5. The computer-implemented method of claim 1 , wherein the convolutional neural network-based classifier batch-wise evaluates the training examples during an epoch.
6. The computer-implemented method of claim 5 , wherein the training examples are randomly sampled into batches, wherein each batch has a predetermined batch size.
7. The computer-implemented method of claim 5 , wherein the convolutional neural network-based classifier iterates evaluation of the training examples over one or more epochs.
8. The computer-implemented method of claim 1 , wherein atrous convolutions conserve partial convolution calculations for reuse as adjacent bases are processed.
9. The computer-implemented method of claim 1 , wherein the atrous convolution rate progresses non-exponentially from a lower residual block group to a higher residual block group.
10. The computer-implemented method of claim 8 , wherein the convolutional neural network-based classifier comprises dimensionality altering layers that reshape spatial and feature dimensions of a preceding input.
11. A non-transitory computer readable storage medium impressed with computer program instructions to construct a convolutional neural network-based classifier for variant classification, the instructions, when executed on a processor, implement a method comprising:
training a convolutional neural network-based classifier, which runs on numerous processors coupled to memory, on training data using a backpropagation-based gradient update technique that progressively matches outputs of the convolutional neural network-based classifier with corresponding ground truth labels;
wherein the convolutional neural network-based classifier comprises groups of residual blocks;
wherein each group of residual blocks is parameterized by a number of convolution filters in the residual blocks, a convolution window size of the residual blocks, and an atrous convolution rate of the residual blocks;
wherein the convolution window size varies between the groups of residual blocks;
wherein the atrous convolution rate varies between the groups of residual blocks;
wherein the training data includes pairs of translated sequences generated from benign variants and pathogenic variants and used as benign training examples and as pathogenic training examples; and
wherein the benign variants include common human missense variants and non-human primate missense variants occurring on alternative non-human primate base triplet sequences that share matching reference base triplet sequences with humans.
12. The non-transitory computer readable storage medium of claim 11 , wherein each residual block comprises at least one batch normalization layer, at least one rectified linear unit (ReLU) layer, at least one atrous convolution layer, and at least one residual connection.
13. The non-transitory computer readable storage medium of claim 11 , wherein the convolutional neural network-based classifier is parameterized by a number of residual blocks, a number of skip connections, and a number of residual connections.
14. The non-transitory computer readable storage medium of claim 11 , wherein atrous convolutions conserve partial convolution calculations for reuse as adjacent bases are processed.
15. The non-transitory computer readable storage medium of claim 11 , wherein the atrous convolution rate progresses non-exponentially from a lower residual block group to a higher residual block group.
16. The non-transitory computer readable storage medium of claim 15 , wherein the convolutional neural network-based classifier comprises dimensionality altering layers that reshape spatial and feature dimensions of a preceding input.
17. A system including one or more processors coupled to memory, the memory loaded with computer instructions to construct a convolutional neural network-based classifier for variant classification, the instructions, when executed on the processors, implement actions comprising:
training a convolutional neural network-based classifier, which runs on numerous processors coupled to memory, on training data using a backpropagation-based gradient update technique that progressively matches outputs of the convolutional neural network-based classifier with corresponding ground truth labels;
wherein the convolutional neural network-based classifier comprises groups of residual blocks;
wherein each group of residual blocks is parameterized by a number of convolution filters in the residual blocks, a convolution window size of the residual blocks, and an atrous convolution rate of the residual blocks;
wherein the convolution window size varies between the groups of residual blocks;
wherein the atrous convolution rate varies between the groups of residual blocks;
wherein the training data includes pairs of translated sequences generated from benign variants and pathogenic variants and used as benign training examples and as pathogenic training examples; and
wherein the benign variants include common human missense variants and non-human primate missense variants occurring on alternative non-human primate base triplet sequences that share matching reference base triplet sequences with humans.
18. The system of claim 17 , wherein each residual block comprises at least one batch normalization layer, at least one rectified linear unit (ReLU) layer, at least one atrous convolution layer, and at least one residual connection.
19. The system of claim 17 , wherein the convolutional neural network-based classifier is parameterized by a number of residual blocks, a number of skip connections, and a number of residual connections.
20. The system of claim 17 , wherein atrous convolutions conserve partial convolution calculations for reuse as adjacent bases are processed.