Accelerating Scientific Discovery with Autonomous Goal-evolving Agents

TL;DR

SAGA tackles the bottleneck of choosing effective objective functions in scientific discovery by introducing a bi-level framework with an outer loop that evolves objectives and an inner loop that optimizes solutions under current objectives. The outer loop is implemented by four agents (Planner, Implementer, Optimizer, Analyzer) and supports three autonomy modes (co-pilot, semi-pilot, autopilot), enabling varying levels of human guidance. Demonstrated across antibiotic design, inorganic materials, functional DNA sequences, and chemical process design, SAGA dynamically builds and refines objectives, mitigating reward hacking and enabling multi-objective trade-offs that align with real-world constraints. The framework shows that automated objective formulation, coupled with flexible human-AI collaboration, can significantly enhance the quality and practicality of discovered solutions, with broad implications for accelerating cross-disciplinary scientific innovation.

Abstract

There has been unprecedented interest in developing agents that expand the boundary of scientific discovery, primarily by optimizing quantitative objective functions specified by scientists. However, for grand challenges in science , these objectives are only imperfect proxies. We argue that automating objective function design is a central, yet unmet requirement for scientific discovery agents. In this work, we introduce the Scientific Autonomous Goal-evolving Agent (SAGA) to amend this challenge. SAGA employs a bi-level architecture in which an outer loop of LLM agents analyzes optimization outcomes, proposes new objectives, and converts them into computable scoring functions, while an inner loop performs solution optimization under the current objectives. This bi-level design enables systematic exploration of the space of objectives and their trade-offs, rather than treating them as fixed inputs. We demonstrate the framework through a broad spectrum of applications, including antibiotic design, inorganic materials design, functional DNA sequence design, and chemical process design, showing that automating objective formulation can substantially improve the effectiveness of scientific discovery agents.

Figures (19)

• Figure 1: The SAGA framework and the example applications. (a) Scientists constantly suffer from reward hacking issues, where optimization agents exploit the approximation error of objective functions and propose undesirable solutions with good scores. (b) Finding optimal objectives that bypass reward hacking issues is difficult due to the large search space of objectives and their relative weights. (c) We propose the SAGA framework to automatically discover optimal objectives and candidate solutions through a bi-level procedure. (d) SAGA operates at three different levels of automation, allowing scientists to steer the objective discovery process in various ways. (e) We apply SAGA to scientific design tasks related to chemistry, biology, and materials. (f) The SAGA framework is capable of implementing different objective functions across disciplines.
• Figure 2: Results for antibiotic design. (a) Comparisons between SAGA and language model baselines. Candidates from all SAGA modes achieve the "drug likeness and activity sweet spot", whereas baselines struggle to balance biological objectives, especially KP activity, with chemical assessments like Drug likeness and Synthesizability. (b) Comparisons across SAGA iterations. Text annotations highlight specific agent feedback on objective evolution that drives the improvement in metric scores across iterations. The solid line means objectives address the evaluation metrics, and the dash line means the metric has not been addressed. (c) Distribution of drug likeness score. Most molecules from SAGA surpass the drug likeness threshold (red dashed line), while AlphaEvolve falls below it, demonstrating its critical misalignment with final objectives. (d) An example of the autopilot feedback loop. The analyzer identifies issues and the planner dynamically evolves objectives, shifting the distribution of the Top 100 Candidate more to the Desired Region of high activity and drug likeness. (e) T-SNE plot of SAGA molecules against known antibiotics. Numbered structures (1-4) are examples of molecules passing all evaluation metrics (\ref{['sup:expdetail_drug']}) that are likely function well in experiments. They contain novel scaffolds distinct from known antibiotic clusters.
• Figure 3: Results for inorganic materials design. (a) Property distributions of generated structures from co-pilot across different iterations and from MatterGen (single) targeting only high magnetic density. (b) Number of stable and novel structures satisfying property requirements found by co-pilot and MatterGen (joint) within 200 DFT property calculations, for targets with magnetic density above 0.2 $\text{\AA}^{-3}$ and HHI score below 1500. It also displays 3D visualizations of two crystal structures proposed by the co-pilot mode that satisfy the design goal. (c) Comparisons between different levels of SAGA and selected baselines on the design task of superhard materials for precision cutting. All evaluation metrics are normalized, with higher scores representing better performance. (d) Comparison of different SAGA modes over three iterations with the same held-out metrics. In each iteration, SAGA analyze the optimized crystal structures, propose new objectives, run property optimization, and select the best candidates across all current iterations. Text annotations highlight specific agent feedback on objective evolution that drives the improvement in metric scores across iterations. The solid line means objectives address the evaluation metrics, and the dash line means the metric has not been addressed. (e) An example of the autopilot feedback loop. SAGA identifies issues and dynamically evolves objectives, successfully proposed novel structures exhibiting high hardness, high elastic modulus, and thermodynamic stability.
• Figure 4: Results for functional sequence design. (a) Comparisons between different levels of SAGA and selected baselines with evaluation metrics (both average score and scaled standard error are reported). Our selected task is to design HepG2 (a cell line of epithelial-like cells from liver) specific enhancers. Our metrics (higher is better). (b) The comparisons of different iterations of two different levels with the same held-out metrics. Each iteration will create new objectives. The solid line means objectives address the evaluation metrics, and the dash line means the metric has not been addressed. (c) An example of SAGA to correct the issues in previous iterations. (d) UMAP visualization of enhancers designed by SAGA, AlphaEvolve, and from references. (e) HepG2-specific motif visualization.
• Figure 5: Results for chemical separation process design. (a) Comparisons between different levels of SAGA and baseline RL (trained to maximize product purity) with evaluation metrics (higher is better). Our selected task is to design process flowsheets for separation of an azeotropic butanol/water mixture with varying feed compositions (between 2%/98%). (b) The comparisons of different iterations of the three SAGA levels ran for three iterations with the same evaluation metrics. In each iteration, SAGA will analyze the processes, create new objectives, run the RL optimization, and select the best candidates across all current iterations. (c) Exemplary designed process by baseline RL agent for separating a 50%/50% butanol-water mixture, demonstrating that maximizing the product purity only leads to full separation but also in applying unit operations that do not have an effect on the separation quality (marked in red), as the RL agent is not penalized for using unnecessary operations. (d) Workflow for using SAGA co-pilot with agent and user actions for two iterations, resulting in optimal process design for separating a 50%/50% butanol-water mixture. (e) Text description of an exemplary chemical process that is used by SAGA.