Search papers, labs, and topics across Lattice.
This study introduces AgentFold, a multi-agent framework that autonomously enhances protein folding models by implementing and validating code modifications through a closed-loop search process. By leveraging a Monte Carlo Tree Search (MCTS) policy, AgentFold efficiently allocates computational resources to explore around 80 model variants, achieving a 7.5% improvement in lDDT over previous approaches within a matched computational budget. The findings reveal critical design patterns that inform future protein folding model development, highlighting the benefits of soft, learnable priors over more disruptive methods.
AgentFold not only outperforms traditional model improvement methods but also uncovers essential design patterns that could redefine protein folding strategies.
Scientific LLM agents have shown promise in literature reasoning, tool use, and experiment planning, but it remains unclear whether they can autonomously improve large, tightly coupled scientific machine-learning systems through executable code changes and computationally expensive validation. We study this question in protein folding, where progress requires coordinated architectural modifications, multi-objective evaluation, and domain-aware interpretation. We present AgentFold, a multi-agent framework that formulates folding-model development as a closed-loop search over executable code variants. Starting from ESMFold, AgentFold proposes hypotheses, implements and debugs code-level modifications, evaluates model variants, analyzes experimental outcomes, and stores both successful and failed interventions in structured memory. An MCTS-style policy allocates computational resources across high-scoring search branches. On an engineering-scale protein-folding codebase comprising more than 2,000 lines of code, AgentFold explores approximately 80 model variants using approximately 5,000 GPU-hours and 170 million LLM tokens. Under a matched computational budget, AgentFold improves the best lDDT by 7.5% over independent Codex proposals and outperforms a random-search control. Beyond model improvement, the resulting intervention traces reveal recurring empirical design patterns: stable gains tend to arise from early, soft, learnable priors and gated refinement, whereas direct geometric perturbations and geometry-conditioned feedback often destabilize training. The code and experimental resources are publicly available at https://github.com/lmqfly/AgentFold.
