cs.LGOct 9, 2026
SaveInference-Time Steering Across Scaffolds and Epitopes for Sample-Efficient De Novo Nanobody Design
Organizations: Alloy Therapeutics Allschwil, Switzerland
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 without sacrificing diversity across distinct therapeutic targets and generative configurations.
Figures & tables
Figure 1: Overview of the inference-time steering and allocation pipeline. (A) Illustration of the resource allocation and steering trajectory progressing from the initial noisy state to the final generated structure . (B) Breakdown of the intermediate resampling step. At timestep , Tweedie’s formula is used to predict the denoised backbone from the noisy sample . The inverse folding model BoltzIF generates sequences for the designed CDRs. These are refolded with ESMFold2; the ipTM is then computed and used to update particle weights, .
Figure 2: Impact of inference-time steering on terminal quality and structural diversity. (A, B, C) Oracle correlation (terminal ESMFold2 vs. terminal AF2-M) across three target-scaffold pairs, with marginal showing steering shifts toward higher ipTM. (D) CDR-H3 structural diversity of passing designs ( RMSD); while steering narrows diversity within a run, it explores distinct structural basins across independent runs. Numbers denote total passing designs. (E) Structural examples of PD-L1 (7eow) candidates showing oracle agreement (top) and divergence (bottom).
Figure 3: Generation of BHRF1 binders across 4 epitope-conditioning approaches. (A) Terminal AF2-M ipTM distributions across baseline, unallocated, and dynamic allocation methods. Dashed lines denote group medians (grey: Vanilla). (B) Budget allocation across targeted epitopes and the unconditional (Hotspot off) baseline ( backbones per method across replicates of ). (C) Designable backbone yield. Segments detail absolute counts and condition-specific success rates. (D) Compute breakdown (GPU-hours) and efficiency (designable backbones per GPU-hour). (E) Structural visualization of targeted BHRF1 epitopes.
| Steered | |||
| Target (Scaffold) | Vanilla | Fixed | ESS |
| Designable / GPU-h | |||
| PD-L1 (7eow) | 10.6 0.9 | 16.9 0.6 | 14.4 0.5 |
| IL-7R (7xl0) | 9.4 0.3 | 17.8 1.2 | 15.3 1.3 |
| BHRF1 (7eow) | 4.8 0.3 | 7.0 2.8 | 6.3 0.8 |
| Pass Rate (%) | |||
Table 1: Compute efficiency and success rate of inference-time steering. Designable backbones per end-to-end GPU-hour and pass rate (%) across three targets.
Figure 4: Dynamic scaffold allocation and the impact of oracle divergence. Generation of PD-L1 binders across six distinct scaffolds. (A) Terminal AF2-M ipTM distributions across baseline, unallocated, and dynamic allocation methods. (B) Budget allocation per scaffold ( backbones per method across replicates of each). (C) Designable backbone yield, detailing absolute counts and scaffold-specific success rates. (D) Compute breakdown (GPU-hours) and efficiency (designable backbones per GPU-hour). (E, F) Oracle correlation (terminal ESMFold2 vs. AF2-M ipTM) for 7eow and 8z8v. (G) Structural overlays showing oracle agreement (7eow) and divergence (8z8v).
Appendix figures & tables8 assets
Supplementary material from the paper’s appendix.
Appendix
| Epitope | Nominated hotspot residues (PDB numbering) |
|---|---|
| Ep1 [ 18 ] | Phe65, Thr74, Glu77, Asp82, Ser85, Arg93 |
| Ep2 [ 10 ] | Val86, Ile90, Arg93 |
| Ep3 | Glu77, His78, Leu81 |
| Hotspot off | none nominated |
Table 2: Epitope-conditioning arms for the BHRF1 (PDB 2WH6 ) experiment.
| Scaffold / Source a | FR Identity b | CDR-H3 Len. | Insertions c |
| 8coh | 85.0 % | 19 | 3 / 3 / 7 |
| sonelokimab | 90.0 % | 17 | 3 / 4 / 4 |
| 7eow | 85.0 % | 21 | 3 / 3 / 7 |
| 7xl0 | 87.5 % | 15 | 3 / 3 / 6 |
| 8z8v | 91.2 % | 8 | 3 / 3 / 3 |
| isecarosmab | 88.8 % | 8 | 3 / 2 / 2 |
Table 3: Properties and generative parameters of the evaluated nanobody scaffolds.
Figure 5: Steered and baseline compute overhead breakdown by generation method and target for three targets. Total GPU-hours expended for 1,152 backbones per target across steering policies. Solid bars represent direct steered/vanilla trajectory generation, while hatched bars account for in-loop and terminal refolding, inverse folding, and post-hoc filtering compute.
Figure 6: Expected designability pass rate vs. MPNN sequence budget (Unallocated). The expected pass rate (y-axis) is the fraction of backbones yielding successful design out of folded sequences (x-axis, ). 1,152 backbone designs per (target, scaffold, method) tuple.
Figure 7: Ablation benchmark for BHRF1 binder generation across epitope-conditioning approaches. (A) Terminal AF2-M ipTM distributions across baselines, dynamic allocation methods, and ablation variants (SMC fixed resampling cadence and SH unsteered). Dashed lines denote group medians (grey: Vanilla). (B) Computational budget allocation across targeted epitopes (Ep1–Ep3) and the unconditional baseline (Hotspot off). Total design budget is held constant at trajectories per method across replicates with designs each. (C) Designable backbone yield evaluated under post-hoc structural validation criteria. Bar segments specify absolute counts and condition-specific success rates per epitope.
Figure 8: CDR-H3 structural diversity for BHRF1 epitope allocation. C RMSD of passing designs across baseline, unallocated steering, and dynamic allocation methods (including SMC and SH ablation variants). Lighter violins denote diversity within a single run, whereas darker violins indicate diversity across multiple runs, highlighting that while steering restricts structural diversity within individual trajectories, it successfully explores diverse productive basins across independent runs. Circles denote the mean, and the numbers at the bottom denote the total passing designs per method.
Figure 9: Ablation benchmark for PD-L1 binder generation and scaffold allocation. Evaluation across six candidate scaffolds (extracted from complex structures 7eow, 8coh, 7xl0, 8z8v, and clinical-stage frameworks isecarosmab and sonelokimab). (A) Terminal AF2-M ipTM distributions across baselines, dynamic allocation methods, and ablation variants (SMC global fixed cadence and SH unsteered). Dashed lines denote group medians (grey: Vanilla). (B) Computational budget allocation per scaffold, holding total trajectory count constant at designs per method across replicates of designs each. (C) Designable backbone yield under post-hoc structural validation criteria. Bar segments specify absolute counts and scaffold-specific success rates.
Figure 10: CDR-H3 structural diversity for PD-L1 Scaffold allocation. C RMSD of passing designs for the 7eow scaffold across baseline, unallocated steering, and dynamic allocation methods (including SMC and SH ablation variants). Lighter violins denote diversity within a single run, whereas darker violins indicate diversity across multiple runs, highlighting that while steering restricts structural diversity within individual trajectories, it successfully explores diverse productive basins across independent runs. Circles denote the mean, and the numbers at the bottom denote the total passing designs per method.