STORM

STORM (Sequence-based Toehold Optimization and Redesign Model) optimizes toehold switch sequences using deep learning to predict sequence-function relationships and redesign nucleic acid sensors for synthetic biology applications.

Key Features:

• Deep Learning Architecture: Implements the Sequence-based Toehold Optimization and Redesign Model (STORM) as a sequence-to-function deep learning framework for toehold switches.
• Gradient Ascent Optimization: Applies gradient ascent on model outputs to redesign and optimize poorly performing toehold switch sequences.
• Training Dataset: Trained on a dataset of 91,534 toehold switches to learn sequence-function relationships.
• Model Interpretation: Examines convolutional filters and saliency maps of nucleic acid sequences to interpret learned features and identify performance-relevant hotspots.
• High-Performing Feature Identification: Identifies sequence motifs and features associated with high-performing toehold switches.
• Redesign of Nucleic Acid Sensors: Generates redesigned toehold switch sequences intended to improve sensor performance.

Scientific Applications:

• Toehold Switch Optimization: Predicts and redesigns programmable toehold switches to improve sensor performance and reduce experimental screening needs.
• Synthetic Circuit Construction: Informs design of optimized toehold-based components for synthetic biology circuits.
• Precision Diagnostics: Supports development of toehold-based diagnostic sensors through predictive sequence optimization.
• Framework for Research: Provides a framework for selection and optimization of toehold switches to guide further synthetic biology studies.

Methodology:

STORM is a sequence-to-function deep learning model trained on 91,534 toehold switches; it uses gradient ascent to optimize sequences and inspects convolutional filters and saliency maps to interpret model features and identify mutation hotspots.