ReCoG: Relational and Compact Context Graph Learning for Few-shot Molecular Property Prediction

Few-shot molecular property prediction (FSMPP) is essential in drug discovery and materials design, where high-quality labeled data are often scarce and expensive to obtain. Despite the promising performance of existing methods, especially context-aware methods, they still face two-fold severe challenges with ${\rm \textit{insufficient structural context modeling}}$ \& ${\rm \textit{redundant auxiliary context learning}}$, leading to inadequate context graph exploration and ineffective information utilization for effective molecule representation learning. To address these, in this paper, we propose a novel framework by learning on ${\rm \mathbf{\underline{Re}}}$ational and ${\rm \mathbf{\underline{C}}}$ompact c${\rm \mathbf{\underline{o}}}$ntext ${\rm \mathbf{\underline{G}}}$raph, named ReCoG, to comprehensively exploit the context graph for expressive molecular property prediction. Specifically, the proposed ReCoG contains two core modules: a **(1) cross-property relational learning module** to better model the structural and relational context information, and a **(2) context graph information bottleneck module** to adaptively suppress irrelevant auxiliary signals for compact context information utilization, followed by a detailed theoretical demonstration regarding the importance of joint relational and compact knowledge extraction in context graphs.

Few-shot molecular property prediction aims to predict important properties of molecules when only a small number of labeled examples are available. This problem is important because obtaining reliable molecular labels often requires costly and time-consuming laboratory experiments, especially in drug discovery and materials design. Existing methods try to use information from related molecules and property tasks, but they may either fail to capture useful relationships or introduce irrelevant auxiliary information. In this work, we propose RECOG, a framework that builds a more informative context graph to model relationships among molecular property tasks and then filters out unnecessary information through a compact learning mechanism. By jointly learning useful relations and removing redundant signals, RECOG can produce more effective molecular representations and improve prediction performance under limited-data settings. This may help researchers screen promising molecules more efficiently before conducting expensive experiments.