Off‑target protein binding is a major source of adverse effects for small‑molecule drugs, yet most structure‑based molecular design methods focus on de‑novo generation of selective compounds rather than improving the selectivity of existing, well‑characterized drugs. We therefore introduce Specificity Optimization (SpecOpt): a task that seeks constrained structural modifications to an existing compound while preserving its molecular identity and drug‑like properties, aiming to increase its binding preference for an intended target over known off‑targets.
To enable systematic evaluation, we build a benchmark from ChEMBL compound‑target interaction data, using curated drug‑mechanism annotations to define intended targets and measured activities to define off‑targets. We then develop an agentic framework that docks each compound against its intended target and off‑targets, extracts residue‑aware atom‑protein contacts, and feeds these differential interactions to a large language model (LLM) to propose targeted structural modifications. Candidates are retained only if they satisfy molecular similarity, ADMET, and docking selectivity criteria between target and off‑targets.
On 915 compounds, the agent improves the target‑off‑target binding gap for 84.8% of cases, shifting the mean gap from ‑0.72 kcal/mol to +0.47 kcal/mol, while maintaining an average Tanimoto similarity of 0.72 to the starting molecules. Ablation studies reveal that residue‑specific contact information is the critical optimization signal: replacing residue identities with binary contact indicators eliminates improvement on all 29 ablation compounds.
These findings establish SpecOpt as a distinct molecular design problem and demonstrate that residue‑aware differential interactions constitute an effective signal for enhancing the specificity of existing compounds.
Review