Combining diffusion and transformer models for enhanced promoter synthesis and strength prediction in deep learning

Area of Science:

• Synthetic biology
• Computational biology
• Bioengineering

Background:

• Engineering synthetic promoters is crucial for optimizing gene expression and metabolic pathways.
• Identifying effective synthetic promoters is challenging due to sequence complexity.
• Deep learning models offer powerful tools for analyzing biological data and designing novel sequences.

Purpose of the Study:

• To develop a diffusion model for designing synthetic promoters in bacteria like Escherichia coli and cyanobacteria.
• To engineer synthetic promoters that mimic natural biological features and enhance transcriptional activity.
• To evaluate the performance of synthetic promoters and predict their strength using a transformer model.

Main Methods:

• Utilized a diffusion model to assimilate biological features from natural promoter sequences for synthetic design.
• Employed a transformer model to assess the efficacy and predict the strength of engineered synthetic promoters.
• Validated the model's effectiveness using datasets from Escherichia coli and cyanobacteria.

Main Results:

• The diffusion model generated synthetic promoters with key biological features similar to natural ones.
• Synthetic promoters from the diffusion model showed greater similarity to natural promoters than those from a variational autoencoder.
• The transformer model achieved higher accuracy in predicting promoter strength compared to a convolutional neural network.

Conclusions:

• Diffusion models are superior for synthetic promoter design compared to other deep learning approaches.
• The integrated platform facilitates the generation and strength prediction of synthetic promoters.
• This work advances the engineering of high-performance synthetic promoters for various biological applications.