Denoising ATAC-Seq Data With Deep Learning

Abstract

The present invention provides methods, systems, computer program products that use deep learning with neural networks to denoise ATAC-seq datasets. The methods, systems, and programs provide for increased efficiency, accuracy, and speed in identifying genomic sites of chromatin accessibility in a wide range of tissue and cell types.

What is claimed is: 
     
         1 . A processor comprising: one or more circuits to use one or more neural networks to identify ATAC-seq information based, at least in part, on a plurality of ATAC-seq counts at each base pair position. 
     
     
         2 . The processor of  claim 1 , wherein the neural network comprises at least three successive convolutional layers followed by a residual connection that sums the ATAC-seq counts at each base pair position input with the final convolutional layer output. 
     
     
         3 . The processor of  claim 2 , wherein each convolutional layer and the residual connection is followed by an ReLU layer. 
     
     
         4 . The processor of  claim 2 , wherein the convolutional layers have a receptive field size of 100 to 10,000. 
     
     
         5 . The processor of  claim 2 , wherein the convolutional layers use a number of filters selected from 1 to 100. 
     
     
         6 . The processor of  claim 2 , wherein the neural network comprises 3 successive convolutional layers of receptive field size 100, 100, and 300, respectively, and using 15, 15, and 1 filters, respectively. 
     
     
         7 . The processor of  claim 1 , wherein the neural networks are trained with suboptimal ATAC-seq information using a loss function to compare neural network output to model ATAC-seq information, and adjusting the neural network until the loss function is minimized. 
     
     
         8 . The processor of  claim 7 , wherein the loss function is a mean square error loss function. 
     
     
         9 . The processor of  claim 7 , wherein the model ATAC-seq information relative to the suboptimal ATAC-seq information comprises at least 4-fold increased number of reads of ATAC-seq counts. 
     
     
         10 . A system comprising: a processor and a memory device, wherein the processor comprises one or more circuits to use one or more neural networks to identify ATAC-seq information based, at least in part, on a plurality of ATAC-seq counts at each base pair position. 
     
     
         11 . The system of  claim 10 , wherein the neural network comprises at least three successive convolutional layers followed by a residual connection that sums the ATAC-seq counts of each segment input with the final convolutional layer output. 
     
     
         12 . The system of  claim 11 , wherein each convolutional layer and the residual connection is followed by an ReLU layer. 
     
     
         13 . The system of  claim 11 , wherein the convolutional layers have a receptive field size of 100 to 10,000. 
     
     
         14 . The system of  claim 11 , wherein the convolutional layers use a number of filters selected from 1 to 100. 
     
     
         15 . The system of  claim 11 , wherein the neural network comprises 3 successive convolutional layers of receptive field size 100, 100, and 300, respectively, and using 15, 15, and 1 filters, respectively. 
     
     
         16 . The system of  claim 10 , wherein the system further comprises: a training engine executable on the processor according to software instructions stored in the memory device, wherein the training engine is configured to: train the neural networks with suboptimal ATAC-seq information using a loss function to compare neural network output to model ATAC-seq information, and adjust the neural network until the loss function is minimized. 
     
     
         17 . A non-transitory computer-readable medium comprising instructions that when executed by a processor of one or more circuits cause the processor to use one or more neural networks to identify ATAC-seq information based, at least in part, on a plurality of ATAC-seq counts at each base pair position. 
     
     
         18 . The non-transitory computer-readable medium of  claim 17 , wherein the medium further comprises instructions for a training engine executable on the processor, wherein the training engine is configured to train the neural networks with suboptimal ATAC-seq information using a loss function to compare neural network output to model ATAC-seq information, and adjust the neural network until the loss function is minimized.