Integrating linear optimization with structural modeling to increase HIV neutralization breadth.

Area of Science:

• Computational biology
• Protein engineering
• Immunology

Background:

• Current computational protein design methods struggle with flexible proteins and large conformational ensembles.
• Existing techniques only sample a limited portion of the sequence space due to energy landscape barriers.

Purpose of the Study:

• To develop a novel computational approach to overcome limitations in protein design sampling methods.
• To enhance the breadth of anti-HIV antibodies against diverse viral strains.

Main Methods:

• Integration of Rosetta software suite for structure-based modeling with machine learning and integer linear programming.
• Development of the BROAD (Breadth-enhancing protein design) method.
• Benchmarking against state-of-the-art multistate design in Rosetta.

Main Results:

• The BROAD method significantly outperforms existing Rosetta methods in protein design.
• Achieved 100% predicted binding against a panel of 180 divergent HIV viral strains for an enhanced anti-HIV antibody (VRC23).
• Identified known binding motifs of broadly neutralizing anti-HIV antibodies within the designed sequences.

Conclusions:

• The BROAD approach effectively enhances the predicted breadth of antibodies.
• The method is generalizable to other protein systems beyond antibodies.
• Predicted variants show greatly increased breadth compared to the wild-type antibody, though in vitro validation is pending.