arXiv：cs.LG· Mason Minot, Simon Friedensohn, C\'edric R. Weber·· 13 小时前
面向样本高效从头纳米抗体设计的跨支架与表位推理时引导
Inference-Time Steering Across Scaffolds and Epitopes for Sample-Efficient De Novo Nanobody Design
AI 导读
研究将生成式纳米抗体设计中的条件选择重构为推理时资源分配问题，提出两种动态预算分配器：联合 Global SMC 公式与基于 Batched Successive Elimination 的解耦方法，可跨配置同步引导并分配算力。实验显示，这些分配器将每 GPU-hour 可设计候选的产出提升近 5 倍，且不牺牲针对不同治疗靶点与生成配置的多样性。
正文
Abstract:Generative molecular design typically relies on over-generation and post-hoc filtering. Inference-time steering methods like Feynman-Kac (FK) steering improve efficiency by biasing trajectories toward high-reward candidates, yet they optimize within a single fixed configuration. In de novo nanobody design, upfront commitment to conditioning choices like scaffold identity or target epitope creates a bottleneck. Since yield varies across these conditions, steering unpromising setups wastes compute. We recast generative condition selection as an inference-time resource-allocation problem. We propose two dynamic budget allocators: a joint Global Sequential Monte Carlo (SMC) formulation that steers and allocates compute across configurations concurrently, and a decoupled approach adapting Batched Successive Elimination (BaSE). Empirically, these dynamic allocators increase the yield of designable candidates per GPU-hour nearly 5$\times$ without sacrificing diversity across distinct therapeutic targets and generative configurations.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.13844 [cs.LG] |
| (or arXiv:2610.13844v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.13844 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mason Minot [view email]
[v1]
Fri, 9 Oct 2026 16:54:05 UTC (2,929 KB)
来源：arXiv：cs.LG · arxiv.org