Deciphering eukaryotic cis-regulatory logic with 100 million random promoters

Abstract

Predicting how transcription factors (TFs) interpret regulatory sequences to control gene expression remains a major challenge. Past studies have primarily focused on native or engineered sequences, and thus remained limited in scale. Here, we use random sequences as an alternative, measuring the expression output of nearly 100 million synthetic yeast promoters comprised of random DNA. Random sequences yield a broad range of reproducible expression levels, indicating that the fortuitous binding sites in random DNA are functional. From this data we learn 'billboard' models of transcriptional regulation that explain 93% of expression variation of test data, recapitulate the organization of native chromatin in yeast, and help refine cis-regulatory motifs. Analyzing the residual variation, we uncover more complex regulatory mechanisms, such as strand, position, and helical face preferences of TFs. Such high-throughput regulatory assays of random DNA provide the large-scale data necessary to learn complex models of cis-regulatory logic.

Massively Parallel Assays and Quantitative Sequence–Function Relationships

“…General studies of cis-regulatory grammar have used a variety of quantitative modeling strategies, including statistical models (149), thermodynamic models (44,98), and neural network models (26,28,129). It remains largely unclear, however, which types of models work best in which situations.…”

Differences between the de novo proteome and its non-functional precursor can result from neutral constraints on its birth process, not necessarily from natural selection alone

“…Since transcription and translation are controlled by local sequence elements, knowledge of these elements may eventually be combined with a random-sequence model to predict the regulatory properties of JPs, like we did for their sequence properties. Studies of the transcriptional activity of synthetic random DNA in Escherichia coli (Yona et al, 2017) and yeast (Boer et al, 2018) show that such sequences frequently contain the patterns required for the initiation and regulation of transcription. Factors that are external to intergenic regions also seem to play a role in the expression of JPs and their functionalization, such as bidirectional promoters (Vakirlis et al, 2017), translated UTRs and translated alternative ORFs within canonical ORFs (Vanderperre et al, 2013).…”

Deciphering regulatory DNA sequences and noncoding genetic variants using neural network models of massively parallel reporter assays

“…Functional genomic assays developed over the last decade (such as ChIP-seq, DNase/ATAC-seq, and others) have allowed for candidate cis-regulatory elements (cCREs) to be mapped on a genome-wide scale in a wide variety of cell lines and tissues 2,3 . They have more recently been supplemented by massively parallel quantitative measurements of the regulatory activity of native cCREs and synthetic constructs in the form of Massively Parallel Reporter Assays (MPRAs) [4][5][6][7][8] and Self-Transcribing Active Regulatory Regions sequencing (STARR-seq) [9][10][11][12][13] as well as direct highthroughput perturbations of cCREs in their native contexts using pooled CRISPR screens 14,15 .…”