Back to feed
arXiv cs.LG·

AgForce Enables Antigen-conditioned Generative Antibody Design

Signal
82
Hype
15
In three linesAgForce, an encoder-decoder architecture with GNN, addresses three failure modes in antibody design: antigen blindness, vocabulary collapse, and inability to generate antigen-specific sequences. Uses framework dropout, gated bottlenecks, hyperbolic attention, and Mixture Density Network. Improves amino acid recovery by 8% on CHIMERA-Bench.
Read source
Your take?
PapersBenchmarksCode generationReinforcement learning

Summary generated by Claude — human-verified