Abstract:Effective transfer learning for molecular property prediction has shown considerable strength in addressing insufficient labeled molecules. Many existing methods either disregard the quantitative relationship between source and target properties, risking negative transfer, or require intensive training on target tasks. To quantify transferability concerning task-relatedness, we propose Principal Gradient-based Measurement (PGM) for transferring molecular property prediction ability. First, we design an optimization-free scheme to calculate a principal gradient for approximating the direction of model optimization on a molecular property prediction dataset. We have analyzed the close connection between the principal gradient and model optimization through mathematical proof. PGM measures the transferability as the distance between the principal gradient obtained from the source dataset and that derived from the target dataset. Then, we perform PGM on various molecular property prediction datasets to build a quantitative transferability map for source dataset selection. Finally, we evaluate PGM on multiple combinations of transfer learning tasks across 12 benchmark molecular property prediction datasets and demonstrate that it can serve as fast and effective guidance to improve the performance of a target task. This work contributes to more efficient discovery of drugs, materials, and catalysts by offering a task-relatedness quantification prior to transfer learning and understanding the relationship between chemical properties. Transfer learning is known to enhance molecular property prediction in limited data sets, however, negative transfer due to insufficient relatedness between source and target tasks continues to be a major challenge. Here, the authors develop a principal gradient-based measurement to evaluate the quantitative transferability from the source property to the target property before applying transfer learning, significantly improving the transfer learning performance.

One-shot heterogeneous transfer learning from calculated crystal structures to experimentally observed materials

Predicting melting temperature of inorganic crystals via crystal graph neural network enhanced by transfer learning

Transfer learning for materials informatics using crystal graph convolutional neural network

Overcoming data scarcity with transfer learning

Application of Deep Transfer Learning to Predicting Crystal Structures of Inorganic Substances

Transfer learning on large datasets for the accurate prediction of material properties

Exploring the potential of transfer learning for metamodels of heterogeneous material deformation

Enhancing crystal structure prediction by combining computational and experimental data via graph networks

Fast and Effective Molecular Property Prediction with Transferability Map

A Universal Foundation Model for Transfer Learning in Molecular Crystals

Transfer Learning in Materials Informatics: structure-property relationships through minimal but highly informative multimodal input

Crystal structure prediction with machine learning-based element substitution

Direct recognition of crystal structures via three-dimensional convolutional neural networks with high accuracy and tolerance to random displacements and missing atoms

Large-scale machine-learning-assisted exploration of the whole materials space

Transfer learning for metal–organic frameworks

Exploring diamond-like lattice thermal conductivity crystals via feature-based transfer learning

Generative Inverse Design of Crystal Structures via Diffusion Models with Transformers

From Bulk Effective Mass to 2D Carrier Mobility Accurate Prediction Via Adversarial Transfer Learning

Crystal structure prediction by combining graph network and optimization algorithm

Two Wrongs Can Make a Right: A Transfer Learning Approach for Chemical Discovery with Chemical Accuracy

Improving neural network predictions of material properties with limited data using transfer learning