Improving Molecular Property Prediction in Small Language Models Using Graph-based Tools

Improving Molecular Property Prediction in Small Language Models Using Graph-based Tools

利用基于图的工具改进小型语言模型中的分子性质预测

Small language models (SLMs) have shown promise for zero-shot molecular property prediction from SMILES strings, yet they often suffer from structural blindness because sequence representations under-specify key graph-topological cues. 小型语言模型(SLMs)在基于 SMILES 字符串的零样本分子性质预测方面展现出了潜力,但它们往往存在“结构盲区”,因为序列表示无法充分体现关键的图拓扑特征。

We propose a modular Context-Augmented Prompting framework that enables agentic tool use at inference time: a trained GNN expert model provides a predictive hint with confidence, and a GNN extracts an instance-specific explanatory subgraph (e.g., a subgraph SMILES and an accompanying explanatory paragraph). 我们提出了一种模块化的“上下文增强提示”(Context-Augmented Prompting)框架,该框架支持在推理阶段使用代理工具:一个经过训练的 GNN(图神经网络)专家模型提供带有置信度的预测提示,同时 GNN 提取特定实例的解释性子图(例如,子图 SMILES 和随附的解释性段落)。

We evaluate three commonly used SLMs on MUTAG and Tox21 under five prompting configurations ranging from SMILES-only to using all available tools at hand. 我们在 MUTAG 和 Tox21 数据集上评估了三种常用的小型语言模型,并采用了五种不同的提示配置,范围从仅使用 SMILES 到使用所有可用工具。

Across two datasets, enriching prompts with graph-derived context yields substantial accuracy gains, often exceeding 25% relative improvement and up to 74% on Tox21. 在两个数据集上,通过图衍生上下文来丰富提示信息,显著提升了预测准确率,相对提升幅度通常超过 25%,在 Tox21 数据集上最高可达 74%。

We further validate the functional relevance of the extracted motifs via a necessity-based edge-drop intervention. 我们进一步通过基于必要性的“边删除干预”(edge-drop intervention)验证了所提取基序的功能相关性。

Despite the observed gains, a persistent gap remains to specialized GNN models, highlighting both the value and limits of text-conditioned reasoning for molecular structure. 尽管取得了上述进展,但与专业的 GNN 模型相比仍存在差距,这凸显了文本条件推理在分子结构分析中的价值与局限性。