Explore deep learning techniques for genomic discovery in this comprehensive workshop presentation from the Novo Nordisk Foundation Center's Multimodal Data Integration event. Delve into high-resolution assays, footprint shapes, and single base pair resolution as Anshul Kundaje from Stanford University's Departments of Genetics and Computer Science guides you through advanced concepts. Learn about linear mapping, additive decomposition, and the intricacies of MODISCO. Discover the importance of TN5 Bias and various bias correction methods, including neural network approaches and genome-wide applications. Gain insights into deep left profiles, motif discovery, and variant effect scores. Engage with the discussion session highlighting key takeaways and exploring the potential of bias factorization in genomic research.
Overview
Syllabus
Introduction
Presentation
high resolution assays
footprint shapes
single base pair resolution
total counts
interpretation
linear mapping
additive decomposition
modisco
how it works
using bodisco
TN5 Bias
Bias corrected profiles
Bias correction with neural network
Bias correction in two ways
Bias correction in actual genomes
Bias correction genomewide
Deep left profiles
Motif discovery
Summary
Discussion session
Highlights
Variant effect score
Bias Factorization
Taught by
Broad Institute
