Artificial neural networks and simulated molecular evolution are potential tools for sequence-oriented protein design

Area of Science:

• Computational biology
• Bioinformatics
• Artificial intelligence in protein design

Background:

• Amino acid sequences contain crucial information for protein function.
• Feature extraction from biological sequences is vital for understanding protein properties.
• Artificial neural networks offer powerful tools for complex pattern recognition.

Purpose of the Study:

• To explore the utility of artificial neural filter systems for feature extraction in amino acid sequences.
• To apply these systems to identify signal peptidase I cleavage sites in protein precursors.
• To evaluate optimization strategies within a simulated molecular evolution framework for protein design.

Main Methods:

• Utilizing trained neural networks as fitness functions in simulated molecular evolution.
• Implementing and comparing gradient search, diffusive search, and evolution strategy optimization schemes.
• Conducting optimization experiments on a multimodal example function to assess performance.

Main Results:

• Artificial neural networks demonstrate potential for feature extraction from amino acid sequences.
• Signal peptidase I cleavage-site analysis serves as a successful example application.
• Evolution strategy proved superior for optimization in high-dimensional, multimodal search spaces compared to gradient search.

Conclusions:

• Trained neural networks are effective for feature extraction and can guide protein design.
• Simulated molecular evolution, powered by neural networks, enables rational, computer-based protein design.
• Evolution strategy is the preferred optimization method for complex, multimodal protein design landscapes.