Constraint-Aware Corrective Memory for Language-Based Drug Discovery Agents
Summary: arXiv:2604.09308v1 Announce Type: new
Abstract
Large language models are making autonomous drug discovery agents increasingly feasible, but reliable success in this setting is not determined by any single action or molecule. It is determined by whether the final returned set jointly satisfies protocol-level requirements such as set size, diversity, binding quality, and developability. This creates a fundamental control problem: the agent plans step by step, while task validity is decided at the level of the whole candidate set. Existing language-based drug discovery systems therefore tend to rely on long raw history and under-specified self-reflection, making failure localization imprecise and planner-facing agent states increasingly noisy.
Introducing CACM
We present CACM (Constraint-Aware Corrective Memory), a language-based drug discovery framework built around precise set-level diagnosis and a concise memory write-back mechanism. CACM introduces protocol auditing and a grounded diagnostician, which jointly analyze multimodal evidence spanning task requirements, pocket context, and candidate-set evidence to:
- Localize protocol violations
- Generate actionable remediation hints
- Bias the next action toward the most relevant correction
Memory Organization
To keep planning context compact, CACM organizes memory into three distinct channels:
- Static: Stores persistent information that does not change frequently.
- Dynamic: Contains data that may vary with each iteration of the planning process.
- Corrective: Focuses on recent failures and corrective actions to improve decision-making.
By compressing these channels before write-back, CACM preserves persistent task information while exposing only the most decision-relevant failures. This innovative approach not only streamlines the agent’s decision-making process but also enhances the overall efficacy of drug discovery efforts.
Experimental Results
Our experimental results indicate that CACM significantly improves the target-level success rate by 36.4% over the state-of-the-art baseline. The findings demonstrate that reliable language-based drug discovery benefits not only from more powerful molecular tools but also from more precise diagnosis and more economical agent states.
Conclusion
In conclusion, CACM represents a pivotal advancement in the field of language-based drug discovery. By addressing the inherent challenges related to task validity and failure localization, this framework provides a more robust and efficient approach for developing autonomous drug discovery agents, ultimately paving the way for more effective therapeutic solutions.