Model-driven generation of artificial yeast promoters

Area of Science:

• Synthetic biology
• Molecular biology
• Computational biology

Background:

• Promoters are crucial for gene regulation but limited sets are commonly used in yeast Saccharomyces cerevisiae.
• Developing predictive models for promoter sequences can accelerate the design of genetic constructs and advance synthetic biology.

Purpose of the Study:

• To generate and utilize models that accurately predict protein expression from promoter sequences.
• To enable rapid generation of useful promoters for Saccharomyces cerevisiae.

Main Methods:

• Measured gene expression activity of over 675,000 constitutive and 327,000 inducible promoter sequences.
• Trained an ensemble of convolutional neural networks on combined constitutive and inducible promoter datasets.

Main Results:

• Achieved high predictive accuracies (R² > 0.79) on multiple sequence-activity prediction tasks.
• Developed model-guided design strategies yielding large, sequence-diverse promoter sets.
• Identified promoters with activities exceeding those in training data and comparable to best-in-class sequences.

Conclusions:

• Model-guided design is a valuable approach for generating functional DNA parts.
• Accurate predictive models facilitate the creation of novel promoters for synthetic biology applications in yeast.