Spectral Manifold Harmonization for Graph Imbalanced Regression

TL;DR

This paper introduces Spectral Manifold Harmonization (SMH), a novel method for imbalanced regression on graph data that generates synthetic samples preserving topology and focusing on relevant target ranges, improving predictive accuracy.

Contribution

SMH is a new approach that addresses the lack of methods targeting domain-specific imbalanced regression on graphs by generating topologically consistent synthetic samples.

Findings

SMH improves predictive performance on chemistry datasets.

SMH effectively focuses on relevant target value regions.

Experimental results show consistent gains over baseline methods.

Abstract

Graph-structured data is ubiquitous in scientific domains, where models often face imbalanced learning settings. In imbalanced regression, domain preferences focus on specific target value ranges that represent the most scientifically valuable cases; however, we observe a significant lack of research regarding this challenge. In this paper, we present Spectral Manifold Harmonization (SMH), a novel approach to address imbalanced regression challenges on graph-structured data by generating synthetic graph samples that preserve topological properties while focusing on the most relevant target distribution regions. Conventional methods fail in this context because they either ignore graph topology in case generation or do not target specific domain ranges, resulting in models biased toward average target values. Experimental results demonstrate the potential of SMH on chemistry and drug discovery benchmark datasets, showing consistent improvements in predictive performance for target domain ranges. Code is available at https://github.com/brendacnogueira/smh-graph-imbalance.git.