Deep Learning for Shallow Sequencing Johnny Israeli Nvidia - PowerPoint PPT Presentation

Deep Learning for Shallow Sequencing Johnny Israeli Nvidia Genomics Group GTC 2018 1

Talk Outline ● A Study of Breast Cancer ● Sequencing Technologies ● Deep Learning for Genomics ● Enhancing Sequencing with Deep Learning 2

100 Pa 2000 Pa 25 μm

ATAC-seq profiling of genomic elements controlling gene expression Map to genome

Control elements activated in stiff cells reveal the SP1 DNA motif

Knock down of SP1 in stiff cells removes invasive phenotype Soft Stiff SP1 Knock down

Sequencing Technologies 8

Genome Analyzer IIx HiSeq 2500 MinION SmidgeION

Sequencers as Counting Devices

Sequencers as Counting Devices

Sequencers as Counting Devices

Sequencers as Counting Devices

Diverse sequencing-based technologies

Growth of single cell sequencing

Deep Learning for Genomics 18

DL4Genomics takes off 2015

DeepSEA (2015): Predicting effects of variations in the non-coding genome with deep learning

DeepVariant (2016): Detection of single nucleotide genomic variations and insertions/deletions with deep learning

Chiron (2017): Translating MinION measurements into DNA Sequence with Deep Learning

1,700 Models!

Enhancing Sequencing with Deep Learning 24

Is this the same ATAC-seq signal?

Is this the same ATAC-seq signal?

Enhancing ATAC-seq with Deep Learning 140- 50M DNA Reads 18- 5M DNA Reads

Enhancing ATAC-seq with Deep Learning 140- 50M DNA Reads 18- 5M DNA Reads 53- 50M DNA Reads 8- 5M DNA Reads

Enhancing ATAC-seq with Deep Learning 140- 50M DNA Reads 18- 5M DNA Reads 53- 50M DNA Reads 8- 5M DNA Reads 53- Linear Model Enhancement

Enhancing ATAC-seq with Deep Learning 140- 50M DNA Reads 18- 5M DNA Reads 53- 50M DNA Reads 8- 5M DNA Reads 53- Linear Model Enhancement 53- Deep Learning Enhancement

Enhancing ATAC-seq with Deep Learning 140- 50M DNA Reads 18- 5M DNA Reads 140- 50M DNA Reads 17- 5M DNA Reads 140- Linear Model Enhancement 140- Deep Learning Enhancement

Enhancing ATAC-seq with Deep Learning 140- 50M DNA Reads 18- 5M DNA Reads 5M DNA Reads Signal Enhanced by DL ~ 20M Signal!

Augmenting Sequencers with Deep Learning Better Data Cheaper Signal More Better Deep Enhancement Sequencing Data Learning Models Generation

Genomics @ Nvidia (we are hiring!) Joint work with Thank you for ● Nvidia AI Infrastructure listening! ● Nvidia Research Questions? ● Nvidia Healthcare Nikolai jisraeli@nvidia.com Yakovenko 34