Abstract
Enhancing accurate molecular property prediction relies on effective and proficient representation learning. It is crucial to incorporate diverse molecular relationships characterized by multi-similarity (self-similarity and relative similarities) (Wang et al., 2019) between molecules. However, current molecular representation learning methods fall short in exploring multi-similarity and often underestimate the complexity of relationships between molecules. Additionally, previous multi-similarity approaches require the specification of positive and negative pairs to attribute distinct pre-defined weights to different relative similarities, which can introduce potential bias. In this work, we introduce Graph Multi-Similarity Learning for Molecular Property Prediction (GraphMSL) framework, along with a novel approach to formulate a generalized multi-similarity metric without the need to define positive and negative pairs. In each of the chemical modality spaces (e.g., molecular depiction image, fingerprint, NMR, and SMILES) under consideration, we first define a self-similarity metric (i.e., similarity between an anchor molecule and another molecule), and then transform it into a generalized multi-similarity metric for the anchor through a pair weighting function. GraphMSL validates the efficacy of the multi-similarity metric across Molecu-leNet datasets. Furthermore, these metrics of all modalities are integrated into a multimodal multi-similarity metric, which showcases the potential to improve the performance. Moreover, the focus of the model can be redirected or cus-tomized by altering the fusion function. Last but not least, GraphMSL proves effective in drug discovery evaluations through post-hoc analyses of the learnt representations.