EVMP: enhancing machine learning models for synthetic promoter strength prediction by Extended Vision Mutant Priority...

Area of Science:

• Synthetic biology
• Metabolic engineering
• Computational biology

Background:

• Accurate prediction of synthetic promoter strength is crucial for metabolic engineering and synthetic biology.
• Experimental annotation of promoter strength is time-consuming and laborious.
• Existing machine learning models for synthetic promoter strength prediction are limited by the proximity of synthetic promoters.

Purpose of the Study:

• To enhance machine learning models for synthetic promoter strength prediction.
• To propose a universal framework, EVMP (Extended Vision Mutant Priority), that utilizes mutation information more effectively.

Main Methods:

• EVMP transforms synthetic promoters into base promoters and k-mer mutations.
• These components are processed by BaseEncoder and VarEncoder, respectively.
• Optional data augmentation generates multiple data copies using different base promoters.

Main Results:

• EVMP enhanced ML model performance in the Trc synthetic promoter library, improving Mean Absolute Error (MAE) by up to 61.30%.
• EVMP improved the state-of-the-art record by 15.25% (MAE) and 4.03% (R²).
• Data augmentation further boosted performance by 17.95% (MAE) and 7.25% (R²) compared to non-EVMP state-of-the-art.

Conclusions:

• Extended vision (k-mer) is essential for EVMP's effectiveness.
• EVMP alleviates the over-smoothing phenomenon, contributing to its predictive power.
• EVMP highlights mutation information, significantly improving synthetic promoter strength prediction accuracy.