A new method to improve how antibodies bind to targets
The iGC framework uses reinforcement learning to optimize antibody affinity maturation.
Researchers developed iGC, which stands for in silico Germinal Centers. This framework optimizes antibody binding through a two-stage process.
The first stage learns a somatic hypermutation (SHM) mutation prior. This prior is learned from parent-child pairs in B-cell lineage trees. The second stage uses reinforcement learning (RL) to optimize antigen binding.
Somatic hypermutation (SHM) is a natural process where B cells acquire mutations. Affinity maturation is a natural optimization process where B cells evolve toward higher affinity. Antigen binding is the process of an antibody connecting to a target.
To stop reward exploitation, the system uses a Reward-Tax Meta-Gradient objective. RewardNet smooths the affinity landscape. TaxNet penalizes reward overexploitation.
The mutation prior more than doubles the strongest baseline in mutation-site prediction. iGC maintains SHM sequence structure after optimization. It also achieves a strong exploration-exploitation balance.
Key facts
- iGC uses a two-stage reinforcement learning framework for antigen-specific antibody optimization.
- The mutation prior more than doubles the strongest baseline in mutation-site prediction.
- iGC maintains SHM sequence structure following RL optimization.