CalibHyper improves few-shot molecular property prediction accuracy
CalibHyper: Chance-Corrected Relational Hypergraphs for Few-Shot Molecular Property Prediction
Machine LearningComputation and Language
Summary
Predicting chemical properties of molecules is important but hard because experiments are expensive and data is limited. The authors propose CalibHyper, a method that better understands how different properties relate by correcting for random chance in their relationships. This method uses advanced math to adjust how it learns connections and shows strong results on multiple datasets for predicting molecular traits even with very few examples. By doing this, CalibHyper helps computers make smarter guesses about molecules when data is scarce.
What this means in practice
- •For drug discovery teams: Enable better predictions of molecular properties to guide early-stage drug candidate evaluation with limited experimental data.
- •For materials science developers: Support informed prediction of material properties from small datasets, accelerating the design of new functional materials.
Authors
Linyu Li, Zhi Jin, Yuanpeng He, Dongming Jin, Huanyu Liu, Huanyao Zhang, Haoran Duan, Heng Tian, Gadeng Luosang, Nyima Tashi
Abstract
Molecular property prediction is central to drug development and materials discovery, but experiments are costly and labeled data are scarce. Context-aware methods use auxiliary assay labels to support few-shot prediction, and recent work supervises property relations with label agreement. However, label agreement is sensitive to class marginals and does not directly capture dependence between properties. We propose CalibHyper, a chance-corrected relational hypergraph method based on the joint label distribution. CalibHyper subtracts an independence baseline from the ordered four-state label distribution and shrinks the residual according to the number of joint observations. A swap-equivariant relation head estimates these residuals, which choose the auxiliary properties for each molecule and set the sign and weight of their hyperedge messages. On thirteen datasets from five benchmarks, in both 1-shot and 10-shot settings, CalibHyper and its ablation settings achieve ROC-AUC competitive with the strongest reported results.