How Computers Predict Protein Shapes Faster Using Smart Templates

This patent describes a computer method to quickly predict the 3D shape of proteins by creating and refining "synthetic templates" from existing protein structures, reducing the heavy computational work usually needed.

What it covers

The patent outlines a multi-step computer method (Claim 1) to predict the 3D structure of a protein, which is an amino acid sequence. It starts by analyzing the protein's sequence to create a "sequence profile matrix" and identify "internal contacts" (Claim 1a-d). These features are then matched ("threaded") against a database of known protein structures, called "original templates" (Claim 1e). The method calculates "normal modes of motion" for these original templates and then "perturbs" them to create many new "synthetic templates" (Claim 1g-h). The system picks the best synthetic templates based on their energy (Claim 1i-j) and combines them with the original ones (Claim 1k). Finally, it runs "Markov Chain Monte Carlo simulations" (Claim 1m) and refines the best resulting shapes through "energy minimization" (Claim 1p) to find the most stable predicted structure (Claim 1q).

What it doesn't cover

• —Protein structure prediction methods that do not rely on a database of existing "original templates."
• —Techniques that do not specifically "perturb" templates using "normal modes of motion" to generate new "synthetic templates."
• —Prediction systems that do not employ "Markov Chain Monte Carlo simulations" for exploring different protein shapes.
• —Methods that skip the final "energy minimization" step to refine the predicted protein models.
• —Protein structure prediction purely based on deep learning or artificial intelligence without the specific template generation and simulation steps outlined.

The clever bit

The novelty lies in creating "synthetic templates" by "perturbing" existing protein structures using their "normal modes of motion." This allows the system to efficiently explore many possible protein shapes without starting from scratch, saving a huge amount of computational time.

Why it matters

Predicting protein structures is crucial for understanding how proteins work and for designing new drugs. Traditional methods are very slow and use a lot of computer power. This patent aims to make that process much faster and more efficient by intelligently creating new template structures, which could speed up drug discovery and biological research significantly.

Real-world examples

• 1.Dnastar Lasergene software suite
• 2.Protein modeling tools in academic research
• 3.Drug discovery platforms
• 4.Biotechnology research software