Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery
Abstract
Despite breakthroughs in AI-driven molecular generation, existing discovery pipelines operate largely in silico, with wet-lab feedback reserved for validation rather than iterative improvement. Whether large language model (LLM)-based reasoning agents can accelerate de novo discovery by iterating with lab data remains unproven. Here we use protein-binding aptamer design as a sandbox (1-2 week cycles, $1-$3 per candidate) to demonstrate a wet lab in a closed loop with AI systems. We tasked agentic pipelines with developing aptamer binders against α-thrombin and myoglobin across three design-test-improve cycles (∼1,000 candidates per round; ∼4,400 constructs). Statistically significant, bead-corrected on-target hit rates among new candidates rose across rounds from 0.0% in Round 1 to 26.0% in Round 2 and 42.1% in Round 3 on α-thrombin (37.1% excluding literature motif descendants), and from 10.2% in Round 1 to 3.7% in Round 2 and 33.5% in Round 3 on myoglobin. Single-clone bead-binding titrations of eight full-length α-thrombin candidates confirmed concordance with pooled sequencing enrichment (Pearson r = 0.90). In these assays, literature-derived motifs engineered by the agents to reduce off-target reactivity in the pooled screen reached apparent affinities of KD = 24.6--$133$ nM, whereas de novo scaffolds showed low-micromolar binding (KD > 1 μM). We also report how our pipelines used myoglobin binding data to exploit the G-quadruplex binder space, diagnose cross-reactivity as on-target specificity dropped from 75.0% to 19.1%, and recover by developing strategies that restored specificity to 72.4%. We demonstrate that empirical lab data is more valuable than existing computational proxies, showing how iterative experimental feedback enables AI agents to diagnose assay artifacts and refine molecular designs across rounds.
Related
- CascadeMAP: Autonomous Closed-loop Optimization of Enzyme Cascades via Microfluidics, Machine Learning and Agentic AI
- Can autonomous Agentic-AI systems accelerate drug discovery and development in the next decade?
- What Transfers Across Protein Pockets? Experience Reuse in Molecular Optimization for Agentic Structure-Based Drug Design
- Special Issue “Molecular Mechanism Discovery of the Bioactive Phytochemicals against Different Diseases Based on Network Pharmacology and Molecular Docking, 2nd Edition”
- Discovery of a molecular glue inhibitor that stabilises a non-productive Kalirin-Rac1 complex
BibTeX
@misc{cutts2026aptamers,
title = {{Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery}},
author = {Zachary Cutts and John Coller and Rachel Chen and Hua Tang and Mathias Bellaiche and Ryder Easterlin and Dina Zverinski and Petar Sirkovic and Avinatan Hassidim and Yossi Matias and Pooja Rao and Annalisa Pawlosky},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.24806},
eprint = {2610.24806},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.24806},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery
AU - Zachary Cutts
AU - John Coller
AU - Rachel Chen
AU - Hua Tang
AU - Mathias Bellaiche
AU - Ryder Easterlin
AU - Dina Zverinski
AU - Petar Sirkovic
AU - Avinatan Hassidim
AU - Yossi Matias
AU - Pooja Rao
AU - Annalisa Pawlosky
PY - 2026
DA - 2026/10/17
PB - arXiv
DO - 10.48550/arxiv.2610.24806
UR - https://arxiv.org/abs/2610.24806
AB - Despite breakthroughs in AI-driven molecular generation, existing discovery pipelines operate largely in silico, with wet-lab feedback reserved for validation rather than iterative improvement. Whether large language model (LLM)-based reasoning agents can accelerate de novo discovery by iterating with lab data remains unproven. Here we use protein-binding aptamer design as a sandbox (1-2 week cycles, $1-$3 per candidate) to demonstrate a wet lab in a closed loop with AI systems. We tasked agentic pipelines with developing aptamer binders against α-thrombin and myoglobin across three design-test-improve cycles (∼1,000 candidates per round; ∼4,400 constructs). Statistically significant, bead-corrected on-target hit rates among new candidates rose across rounds from 0.0% in Round 1 to 26.0% in Round 2 and 42.1% in Round 3 on α-thrombin (37.1% excluding literature motif descendants), and from 10.2% in Round 1 to 3.7% in Round 2 and 33.5% in Round 3 on myoglobin. Single-clone bead-binding titrations of eight full-length α-thrombin candidates confirmed concordance with pooled sequencing enrichment (Pearson r = 0.90). In these assays, literature-derived motifs engineered by the agents to reduce off-target reactivity in the pooled screen reached apparent affinities of K_D = 24.6--$133$ nM, whereas de novo scaffolds showed low-micromolar binding (K_D > 1 μM). We also report how our pipelines used myoglobin binding data to exploit the G-quadruplex binder space, diagnose cross-reactivity as on-target specificity dropped from 75.0% to 19.1%, and recover by developing strategies that restored specificity to 72.4%. We demonstrate that empirical lab data is more valuable than existing computational proxies, showing how iterative experimental feedback enables AI agents to diagnose assay artifacts and refine molecular designs across rounds.
ER -