Conditional generation of antibody sequences with classifier-guided germline-absorbing discrete diffusion

Short summary

Researchers introduce germline-absorbing discrete diffusion, a machine learning method for computationally designing therapeutic antibodies that learns meaningful biological variation rather than memorizing germline sequences. By using the germline sequence as the diffusion absorbing state, the approach improves non-germline residue prediction accuracy from 26% to 46%, approaching theoretical biological limits. The method outperforms existing strategies on conditional generation tasks like optimizing hydrophobicity and predicted binding affinity.

• •Novel germline-absorbing discrete diffusion method addresses antibody design limitations
• •Improves non-germline residue prediction accuracy from 26% to 46%
• •Outperforms EvoProtGrad on conditional generation for hydrophobicity and binding affinity optimization

Generated with AI, which can make mistakes.