Project Summary In each of our trillions of cells, genetic information is stored on meter-long chromosomes spatially organized inside micron-scale nuclei. In the past decade, large-scale efforts have built increasingly detailed atlases of transcription, regulatory elements, and genome folding across ever-expanding sets of cell types and tissues. Still, these beautiful maps do not by themselves reveal the sequences or molecular mechanisms acting in these diverse cellular contexts. Drawing on approaches from biophysics, bioinformatics, and machine learning, this project will develop novel computational approaches to determine the DNA sequences and mechanisms underlying 3D genome organization, and how this in turn relates to genomic functions. Drawing on the latest breakthroughs in machine learning, we will model how individual nucleotides contribute to genome folding. We will apply these models to characterize cell-type specific genome folding, develop methods to engineer DNA sequences in silico, and model enhancer-promoter influences. Concurrently, we will build biophysical models to understand deeply conserved mechanisms of genome folding. Using meiotic chromosome folding as a model system, we will develop models to learn new rules governing cohesin dynamics and loop extrusion. To uncover how extrusion interfaces with other mechanisms, we will build models of synaptonemal complex assembly, as well as models of meiotic chromosome organization across species. By honing in on the sequences most crucial for locus-specific genome folding and characterizing the mechanisms driving genome- wide folding, the computational models we develop and the mechanisms we discover will enable new approaches to precision genome engineering. This will include how to re-wire gene-regulatory circuits, not only by targeting enhancers and promoters, but also by modulating their cell-type specific communication. At the same time, the aims described here will bridge fundamental insights into 3D chromosome organization with clinical genomics, and greatly improve the interpretability of non-coding DNA variants.