[email protected] luke harries neural networks …...mutations (germline variant calling) • uses...

28
Neural Networks for Genomic Variant Calling Luke Harries [email protected] * Circos Plot visualization of tumour WGS.

Upload: others

Post on 25-May-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Neural Networks for Genomic Variant CallingLuke [email protected]

* Circos Plot visualization of tumour WGS.

Page 2: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Agenda

• Why detect cancer mutations?• GermlineNet - Detecting inherited mutations• Spliced kernel• SomaticNet - Detecting cancer mutations

Page 3: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Why detecting cancer mutations? (Somatic Variant Calling)

• Cancer is caused by the progressive accumulation of DNA mutations

• Knowing which mutations have occurred allows doctors to:• Target the cancer with

chemotherapy• Detect resistance to treatment

up to seven months earlier than scans

https://www2.le.ac.uk/projects/vgec/highereducation/topics/cellcycle-mitosis-meiosis

Page 4: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

••

Page 5: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Next-Generation Sequencing

Sequencing DNA has a high error rate: 0.1-10%Fox et al. 2014https://www.researchgate.net/figure/Basic-principle-of-next-generation-sequencing-technologies_fig4_291171327

Page 6: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Output of the sequencers

Page 7: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Detecting cancer mutations (Somatic Variant Calling)

Page 8: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Existing Variant Callers

• Low concordance rate (based on different heuristics)• False assumption that read errors are independent

Cai et al 2016

Page 9: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

DeepVariant

• Highest performance at a related task - detecting inherited mutations (Germline Variant Calling)

• Uses a deep convolutional neural network on encoded pileup images

Page 10: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Detecting Inherited Mutations (Germline Variant Calling)

Page 11: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 12: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

He et al (2015)

Results for 4-fold cross-validation

Page 13: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 14: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 15: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 16: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 17: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 18: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited
Page 19: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Results for 4-fold cross-validation

Page 20: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Detecting cancer mutations - (Somatic Variant Calling)

Page 21: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Siamese Deep Convolutional Network

Omniglot Verification

Facial Verification

Page 22: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

SomaticNet - Deep learning based somatic variant caller

Page 23: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

SomaticNet

Results for 4-fold cross-validation

Page 24: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Summary

• Developed GermlineNet - a deep learning based germline variant caller, inspired by Google’s DeepVariant

• Improved GermlineNet by introducing a novel kernel design - the spliced kernel

• Developed SomaticNet - a novel approach to somatic variant calling which uses a Siamese deep convolution neural network

Page 25: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Thank you!Luke [email protected]

* Circos Plot visualization of tumour WGS.

Page 26: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

References

• Ryan Poplin, Pi-Chuan Chang, David Alexander, Scott Schwartz, Thomas Colthurst, Alexander Ku, Dan Newburger, Jojo Dijamco, Nam Nguyen, Pegah T. Afshar, Sam S. Gross, Lizzie Dorfman, Cory Y. McLean, and Mark A. DePristo. Creating a universal SNP and small indel variant caller with deep neural networks. bioRxiv, page 092890, 3 2018. doi: 10.1101/092890. URL https://www.biorxiv.org/content/early/2018/03/20/ 092890.

• Gregory Koch, Richard Zemel, and Ruslan Salakhutdinov. Siamese Neural Networks for One-shot Image Recognition. Proceedings of the 32nd International Conference on Machine Learning, 2015. URL https://www.cs.cmu.edu/~rsalakhu/papers/oneshot1.pdf.

• Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the Inception Architecture for Computer Vision. 12 2015. URL http://arxiv.org/abs/1512.00567.

• Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep Residual Learning for Image Recognition. 12 2015. URL http://arxiv.org/abs/1512.03385.

• Lei Cai, Wei Yuan, Zhou Zhang, Lin He, and Kuo-Chen Chou. In-depth comparison of somatic point mutation callers based on different tumor next-generation sequencing depth data. Scientific Reports, 6(1):36540, 12 2016. ISSN 2045-2322. doi: 10.1038/ srep36540. URL http://www.nature.com/articles/srep36540.

Page 27: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

Inception v4

Szegedy et al 2016

Page 28: luke.harries@me.com Luke Harries Neural Networks …...mutations (Germline Variant Calling) • Uses a deep convolutional neural network on encoded pileup images Detecting Inherited

ResNet34

He et al (2015)