Related Experiment Video
Updated: Jan 14, 2026
Design and Implementation of an Automated Illuminating, Culturing, and Sampling System for Microbial Optogenetic Applications
Published on: February 19, 2017
Directed optimization and generation of yeast promoter sequences driven by deep learning
Chong Wang1, Shuxin Li2, Xiao Guo3
1School of Medical Engineering, Henan Medical University, Xinxiang, 453000, China; Engineering Technology Research Center of Neurosense and Control of Henan Province, China; Henan International Joint Laboratory of Neural Information Analysis and Drug Intelligent Design, Xinxiang, 453000, China.
DOSDiff, a new diffusion model, enables precise promoter engineering by learning gene sequence rules. It optimizes gene expression without needing pre-trained models, showing success in yeast species.
Area of Science:
- Synthetic biology
- Computational biology
- Molecular biology
Background:
- Precise control of gene expression is crucial for biological applications.
- Current promoter design methods are limited by incomplete understanding of sequence features and reliance on sequence-to-expression predictors.
- There is a need for novel computational frameworks for efficient and accurate promoter engineering.
Purpose of the Study:
- To develop a diffusion-based framework, DOSDiff, for learning promoter sequence features and enabling targeted optimization.
- To achieve promoter optimization through local sequence editing without requiring a pre-trained sequence-to-expression model.
- To demonstrate the effectiveness and generalizability of DOSDiff across different yeast species.
Main Methods:
- Developed DOSDiff, a diffusion-based framework for promoter design and optimization.
- Employed local sequence editing for targeted promoter modification.
- Validated promoter generation by assessing 4-mers distribution similarity.
- Performed in vivo validation of engineered promoters in Saccharomyces cerevisiae and Pichia pastoris.
Main Results:
- DOSDiff achieved a 4-mers distribution similarity of 0.8910 ± 0.0002, significantly outperforming WGAN-GP and DBGM.
- Locally optimized genes in Saccharomyces cerevisiae showed one-to-one expression enhancement.
- Optimized Pichia pastoris promoters retained functional activity and achieved up to 1.70-fold expression gain in vivo.
- DOSDiff demonstrated generalization capabilities across different yeast species.
Conclusions:
- DOSDiff provides a powerful and adaptable framework for precise promoter engineering.
- The method overcomes limitations of existing approaches by learning promoter encoding rules directly from sequence data.
- DOSDiff facilitates targeted gene expression control and holds promise for various synthetic biology applications.
