BRS: Carl de Boer "The evolution of cis-regulatory sequences (in yeast)"

Zoom Recording below:
Topic: Dr de Boer :  The evolution of cis-regulatory sequences
Start Time : Sep 23, 2020 11:41 AM
Meeting Recording:
https://ubc.zoom.us/rec/share/OG7ToIBBc5xpbzLcPYbar67ET3YFUMzyN-jNahUhX…
Contact Katie Beall for passcode (brcadmin@biodiversity.ubc.ca) 

host: Judith Mank

Abstract: Promoters are non-coding DNA sequences that regulate a gene’s expression. Mutations in promoters can alter gene expression, and organismal phenotype and fitness.  A complete fitness landscape, relating DNA sequence to organismal fitness, is a central goal of molecular evolution, but has remained elusive in part from the limited ability of models to generalize across the vast DNA sequence space. We show that by abstracting away the gene- and environment-specific expression-fitness relationship, we can focus on the much more tractable problem of creating a “complete expression landscape”. We approximate the “complete expression landscape” with a deep convolutional neural network trained on high-throughput reporter assay data (promoters randomly sampled from the complete sequence space). Since this model can predict expression with high fidelity on arbitrary DNA sequences, we use it to characterize the regulatory fitness landscape. We find that expression is highly plastic, with expression having the potential to change dramatically with few mutations, show how regulatory complexity arises, identify genes whose expression is under positive and negative selection, and characterize how selective pressures shape sequence divergence.  Finally, we use an autoencoder to capture the promoter sequence “mutational landscape”; here, sequences fall within four mutational archetypes capturing mutational robustness-vs-plasticity, and high-vs-low expression level. Overall, this framework for navigating the complete promoter fitness landscape will help us to understand cis-regulatory evolution.