Sequence design for three-dimensional genome folding using Akita Semifreddo
Smaruj, P. N.; Kelley, D. R.; Fudenberg, G.
Show abstract
Mammalian genomes display complex three-dimensional organization which is crucial for downstream processes like gene regulation. Local features of genome organization are largely driven by loop extrusion and manifest as boundaries, dots, and flames in genome contact maps. Still, the rational design of DNA sequences that produce desired folding patterns has not been demonstrated. Here, we present Akita Semifreddo, a framework that enables the rational in silico design of DNA sequences with programmable 3D folding outcomes. This combines a computationally efficient "half-frozen" version of the AkitaV2 genome folding model with the Ledidi sequence optimizer. We systematically demonstrate that this approach spans the full repertoire of known local folding features. We show that [~]2 kb synthetic sequences can be designed to induce boundaries, dots, and flames at desired strengths, with CTCF motif configurations consistent with their known mechanistic bases. We further demonstrate that weak boundaries can be designed through transcription-associated sequence features alone, without introducing CTCF motifs, and that strong boundaries can be suppressed by introducing SINE B2 retroelement-like sequences. Collectively, these results reveal a many-to-one relationship between DNA sequence and folding outcomes and uncover the biological basis of sequence features leveraged by our model. In short, Akita Semifreddo provides a platform for dissecting the sequence grammar of three-dimensional chromatin architecture and engineering synthetic regulatory landscapes.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Epiphany: predicting Hi-C contact maps from 1D epigenomic signals 97%
- Towards In-Silico CLIP-seq: Predicting Protein-RNA Interaction via Sequence-to-Signal Learning 96%
- Integrative epigenomic and functional characterization assay based annotation of regulatory activity across diverse human cell types 96%
Similar papers in this journal
- Connecting high-resolution 3D chromatin organization with epigenomics 96%
- Hi-C-LSTM: Learning representations of chromatin contacts using a recurrent neural network identifies genomic drivers of conformation 96%
- Normalisr: normalization and association testing for single-cell CRISPR screen and co-expression 96%
Similar papers in this journal
- scTIE: data integration and inference of gene regulation using single-cell temporal multimodal data 95%
- Decoil: Reconstructing extrachromosomal DNA structural heterogeneity from long-read sequencing data 95%
- CoRAL accurately resolves extrachromosomal DNA genomestructures with long-read sequencing 95%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.