On the Scalability of GNNs for Molecular Graphs

TL;DR

This paper demonstrates that scaling GNNs in depth, width, and data significantly improves their performance on molecular graph tasks, leading to a new foundation model that outperforms previous methods in drug discovery applications.

Contribution

It is the first comprehensive study showing the benefits of scaling GNNs for molecular graphs and introduces MolGPS, a new large-scale graph foundation model.

Findings

GNNs benefit greatly from increased scale in various dimensions.

MolGPS outperforms previous models on 26 of 38 downstream tasks.

Strong finetuning scaling behavior observed across multiple tasks.

Abstract

Scaling deep learning models has been at the heart of recent revolutions in language modelling and image generation. Practitioners have observed a strong relationship between model size, dataset size, and performance. However, structure-based architectures such as Graph Neural Networks (GNNs) are yet to show the benefits of scale mainly due to the lower efficiency of sparse operations, large data requirements, and lack of clarity about the effectiveness of various architectures. We address this drawback of GNNs by studying their scaling behavior. Specifically, we analyze message-passing networks, graph Transformers, and hybrid architectures on the largest public collection of 2D molecular graphs. For the first time, we observe that GNNs benefit tremendously from the increasing scale of depth, width, number of molecules, number of labels, and the diversity in the pretraining datasets. We further demonstrate strong finetuning scaling behavior on 38 highly competitive downstream tasks, outclassing previous large models. This gives rise to MolGPS, a new graph foundation model that allows to navigate the chemical space, outperforming the previous state-of-the-arts on 26 out the 38 downstream tasks. We hope that our work paves the way for an era where foundational GNNs drive pharmaceutical drug discovery.