TIMED-Design

TIMED-Design applies Convolutional Neural Networks (CNNs) to perform flexible and efficient protein sequence design by encoding proteins as three-dimensional voxel grids that allow incorporation of design constraints.

Key Features:

• Convolutional Neural Networks (CNNs): Uses CNNs as the core modeling approach for protein sequence design.
• Three-dimensional voxel representation: Represents proteins within a three-dimensional voxel grid for input to the CNN models.
• Design constraint embedding: Allows additional design constraints to be incorporated directly into the input data.
• Reimplementation of CNN models: Includes reimplementations of previously described CNN models as part of development.
• Benchmarking against physics-based methods: Demonstrates substantially reduced computational cost relative to physics-based methods while maintaining or surpassing their performance.

Scientific Applications:

• Protein design and engineering: Designing and engineering protein sequences with specified constraints.
• Synthetic biology: Generating sequence variants for synthetic biology applications requiring novel or optimized proteins.
• Drug discovery: Exploring protein sequence space rapidly to support drug discovery efforts.

Methodology:

Applies Convolutional Neural Networks trained on protein representations encoded as three-dimensional voxel grids with additional design constraints embedded in the input; includes reimplementations of previously described CNN models and benchmarking against physics-based methods.