AI Tool Designs DNA Sequences to Fold Genomes in Desired 3D Patterns

Researchers have introduced Akita Semifreddo, a computational framework that designs DNA sequences intended to produce specified three-dimensional genome folding patterns. The method, described in Genome Biology by Paulina N. Smaruj, David R. Kelley, and Geoffrey Fudenberg, pairs an efficient “half-frozen” version of the AkitaV2 genome folding model with the Ledidi sequence optimizer, allowing DNA sequences with desired folding behavior to be designed entirely in silico.

DNA sequence directs genome folding, which in turn shapes enhancer-promoter communication and gene expression. According to the authors, how to design sequences that yield desired folding patterns had remained largely unexplored. Akita Semifreddo addresses this gap by making the folding model fast enough to sit inside an iterative optimization loop, in which candidate sequences are repeatedly evaluated and revised toward a target folding outcome.

In demonstrations reported in the paper, the framework enabled design for major folding features — boundaries, dots, and flames — with desired positions and strengths. These feature classes correspond to structural elements of genome organization that the underlying AkitaV2 model can predict from sequence, and the design pipeline was able to steer sequences toward specified configurations of each.

The study’s results are computational: the authors present in silico design demonstrations rather than experimental validation of synthesized sequences in cells. The findings therefore establish a design capability within the model’s predictions, and their behavior in laboratory settings remains to be tested.

Beyond design, the researchers position Akita Semifreddo as a platform for two broader purposes: dissecting the sequence grammar of genome folding — that is, learning which sequence features drive which folding outcomes — and engineering synthetic regulatory landscapes. By making folding objectives directly optimizable, the tool offers a way to probe how sequence encodes spatial genome organization and, potentially, to construct regulatory environments with prescribed folding architectures.

Subject of Research: Computational design of DNA sequences for three-dimensional genome folding

Article Title: Sequence design for three-dimensional genome folding using Akita Semifreddo

Article References: Smaruj, P. N., Kelley, D. R., & Fudenberg, G. (2026). Sequence design for three-dimensional genome folding using Akita Semifreddo. Genome Biology. https://doi.org/10.1186/s13059-026-04313-6

Image Credits: AI Generated

DOI: 10.1186/s13059-026-04313-6

Keywords: genomics, 3D genome folding, deep learning, sequence design, computational biology, synthetic biology, Sequence, design, three-dimensional, genome, folding, Akita