External Seminar Leopold Parts "Expanding, minimizing, and randomizing the human genome"

Speaker: Leopold Parts - Wellcome Sanger Institute, UK

Title: Expanding, minimizing, and randomizing the human genome

Abstract: The human genome reference sequence is a quarter century old, and the ongoing challenge is to interpret the variation in it. While predicting structures and mutation effects for protein coding genes is making rapid strides, there remains a gap in our understanding of the remaining 99% that is non-coding. This gap is due to the number of nucleotides to measure in an assay, and the rarity of signals outside exons. I will talk about three projects to address this gap. First, we used a chromosome transfer strain with extensive genomic characterization to discover how 200Mb of evolutionarily distant sequence function in a human cell. These data also allowed us to probe the limits of current AI approaches - where do the huge models succeed and fail in predicting the output of a chromosome in a cell? Second, we attempted to generate qualitatively novel types of data at scale to address the gaps in models’ abilities to account for distal gene regulation. To do so, we created randomized regulatory regions by inserting recombinase sites around enhancer elements, and induced rearrangements with the recombinase. We applied our approach to dissect a distal super-enhancer of the OTX2 gene, generating over 100 alternative regulatory configurations in a single experiment, and establishing how they drive gene expression and chromatin accessibility as well as the individual contributions of its elements to this activity. Third, we used paired prime editing in pooled screens to delete multiple kilobases of sequence in a massively parallel way. We discovered hundreds of non-coding regions with fitness effects, and validated deletions with impact as well as tolerated ones in regions with functional annotations. Together, these approaches can scale to shed light on the non-coding genome dispensability, resolve many human gene regulatory landscapes, and clarify the limitations of the current predictability of genome function, ultimately paving the way to writing desired functions directly into DNA at scale.

Invited by Daan Noordermeer