Automated antibody structure prediction using Accelrys tools: results and best practices

Area of Science:

• Structural biology
• Computational chemistry
• Immunoinformatics

Background:

• Antibody structure prediction is crucial for therapeutic development.
• The Antibody Modeling Assessment experiment provides a benchmark for prediction methods.
• Accurate modeling of antibody Fv fragments is essential for understanding antigen binding.

Purpose of the Study:

• To evaluate template-based modeling strategies for antibody Fv fragment structure prediction.
• To assess the accuracy of predicted antibody structures in the second Antibody Modeling Assessment experiment.
• To identify challenges and areas for improvement in antibody modeling techniques.

Main Methods:

• Template-based modeling using sequence similarity for framework regions.
• Single, chimeric, or multiple template approaches based on template quality.
• Grafting hypervariable loop regions from templates.
• Ab initio refinement for the H3 loop region.
• Constrained energy minimization for final model refinement.

Main Results:

• Accurate models for antibody framework and canonical CDR regions were constructed, with RMSDs below 1 Å on average.
• Prediction of the H3 hypervariable loops remains a significant challenge.
• Submitted models demonstrated high quality, with local geometry assessment scores comparable to experimental structures.

Conclusions:

• Template-based modeling, combined with ab initio refinement and energy minimization, yields accurate antibody framework and CDR models.
• The H3 loop region presents a persistent challenge in antibody structure prediction.
• The evaluated methods and tools are capable of producing high-quality antibody models suitable for further analysis.