# Assessing zero-shot generalisation behaviour in graph-neural-network interatomic potentials

## TL;DR

This paper tests how well a machine learning model trained on graphene oxide can predict properties of unrelated molecules and reactions.

## Contribution

The study quantifies zero-shot generalization of a graph-based MLIP model across different chemical domains.

## Key findings

- GO-MACE-23 shows limited zero-shot performance for small molecules and reactions outside its training domain.
- The model's generalization ability is explored in both molecular and reaction prediction tasks.
- Findings highlight the need for improved transferability in foundational MLIP models.

## Abstract

With the rapidly growing availability of machine-learned interatomic potential (MLIP) models for chemistry, much current research focuses on the development of generally applicable and “foundational” MLIPs. An important question in this context is whether, and how well, such models can transfer from one application domain to another. Here, we assess this transferability for an MLIP model at the interface of materials and molecular chemistry. Specifically, we study GO-MACE-23, a model designed for the extended covalent network of graphene oxide, and quantify its zero-shot performance for small, isolated molecules outside its direct scope, as well as for examples of chemical reactions. Our work provides quantitative insight into the generalisation ability of graph-based MLIP models and, by exploring their limits, can help to inform future developments.

We explore to what extent a machine-learned interatomic potential trained for graphene oxide is applicable to isolated molecules and reactions in a ‘zero-shot’ setting.