Structurally-aware Genetic Algorithm for Vaccine Development

ICLR 2026 Conference Submission20205 Authors

Keywords: Structurally-inspired genetic algorithms, Vaccines development

Abstract: While vaccines are a staple of modern medicine, the development of new vaccines may require decades and billions of dollars. AI-powered protein design has led to a number of promising tools to accelerate this process. Among those, genetic algorithms (GA) specialized for antigen engineering have been used to triage a large number of designs in early steps of the vaccine development process. However, existing approaches typically treat the design process as a simple string optimization problem (representing amino acid sequences), without considering the three-dimensional structure which influences protein function and interactions in real biological systems. A family of structure-aware GAs is proposed that explicitly incorporate protein structural information into evolutionary operators. Rather than exchanging arbitrary sequence segments, amino acids are partitioned into functionally meaningful groups based on structural relationships. The methods are evaluated in silico with a real SARS-CoV-2 spike protein optimization problem, and significant improvements over the traditional GA approach are shown both in performance of found designs and diversity. The results demonstrate that incorporating structural knowledge into evolutionary search can provide solid gains in exploring the design space in early steps of computational vaccine design.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 20205