Combinatorial Optimization of Antibody Libraries via Constrained Integer Programming

Designing effective antibody libraries is a challenging combinatorial search problem in computational biology. We propose ProtLib- Designer (PLD), a constrained integer linear programming (ILP) method that generates libraries while explicitly controlling library size and diversity. PLD uses in silico deep mutational scanning scores from protein language models and antibody inverse-folding models to guide optimization in a cold-start setting, without requiring target-specific experimental data. Across three antibody– antigensystems(Trastuzumab/HER2, D44. 1/HEL, andSpesolimab/ IL36R), PLDproduceslibrariesthatimprovepredictedbinding-related objectiveswhileenforcinguser-specifieddiversityconstraints. Code is available at github. com/LLNL/protlib-designer.

Authors

• Conor F. Hayes Cognizant AI Lab
• Andre R Goncalves Lawrence Livermore National Labs
• Steven Magana-Zook Lawrence Livermore National Labs
• Jacob Pettit Lawrence Livermore National Labs
• Ahmet Can Solak Lawrence Livermore National Labs
• Daniel Faissol Lawrence Livermore National Labs
• Mikel Landajuela Lawrence Livermore National Labs

Keywords

• Combinatorial Optimization
• Integer Programming
• Antibody Library Design
• Diversity Constraints
• Protein Language Models
• Inverse Folding