Purely Agentic Black-Box Optimization for Biological Design

TL;DR

PABLO reframes biological design as a purely agentic, language-based optimization process driven by a hierarchy of LLM agents. By distilling data into a global context and separating global exploration (Explorer) from local strategy design (Planner) and execution (Worker), the framework avoids handcrafted acquisition functions and adapts strategies online. Empirical results show state-of-the-art performance on GuacaMol and antimicrobial peptide design, with improved sample efficiency and competitive token usage; extensions like retrieval-augmented literature and task awareness further boost performance. In vitro validation of PABLO-optimized peptides demonstrates practical potential for therapeutic discovery, while the approach generalizes across modalities and constraints, offering a flexible, knowledge-grounded pathway for complex biological design tasks.

Abstract

Many key challenges in biological design-such as small-molecule drug discovery, antimicrobial peptide development, and protein engineering-can be framed as black-box optimization over vast, complex structured spaces. Existing methods rely mainly on raw structural data and struggle to exploit the rich scientific literature. While large language models (LLMs) have been added to these pipelines, they have been confined to narrow roles within structure-centered optimizers. We instead cast biological black-box optimization as a fully agentic, language-based reasoning process. We introduce Purely Agentic BLack-box Optimization (PABLO), a hierarchical agentic system that uses scientific LLMs pretrained on chemistry and biology literature to generate and iteratively refine biological candidates. On both the standard GuacaMol molecular design and antimicrobial peptide optimization tasks, PABLO achieves state-of-the-art performance, substantially improving sample efficiency and final objective values over established baselines. Compared to prior optimization methods that incorporate LLMs, PABLO achieves competitive token usage per run despite relying on LLMs throughout the optimization loop. Beyond raw performance, the agentic formulation offers key advantages for realistic design: it naturally incorporates semantic task descriptions, retrieval-augmented domain knowledge, and complex constraints. In follow-up in vitro validation, PABLO-optimized peptides showed strong activity against drug-resistant pathogens, underscoring the practical potential of PABLO for therapeutic discovery.

Figures (12)

• Figure 1: A graphical overview of one optimization iteration of PABLO. Each iteration begins with (i) global candidate exploration, (ii) strategy generation via the Planner Agent, and (iii) local refinement of incumbents via planner-proposed strategies. All candidate generations are filtered for validity, novelty, and feasibility before evaluation. PABLO-pseudocode is also provided in \ref{['alg:agent_opt']}.
• Figure 2: Comparing the average number of LLM tokens used per run by PABLO and other LLM-based baselines.
• Figure 3: GuacaMol optimization results on 10 tasks. Curves show objective value of the best molecule found so far as a function of black-box evaluations. PABLO achieves state-of-the-art performance, rapidly reaching strong objective values across tasks.
• Figure 4: PABLO ablation on representative GuacaMol tasks showing the contribution of the Planner and Explorer Agents.
• Figure 5: Antimicrobial peptide (AMP) optimization. We plot predicted MIC versus black-box evaluations (lower MIC is better). (Upper Left): The predicted MIC of the best peptide found so far. (Lower Left): Template-constrained optimization; we show the predicted MIC of the best "feasible" peptide found so far. (Upper Right): Template-free optimization of a diverse portfolio of 20 AMPs; we show the mean predicted MIC of the best sufficiently-diverse portfolio so far.