# Performance of Meta’s Universal Model for Atoms across the Conformational and Configurational Space of Diverse Transition-Metal Catalysts

## TL;DR

This paper evaluates a machine learning model for predicting transition-metal catalyst properties, showing it works well for some systems but needs expert validation for others.

## Contribution

The study introduces a new evaluation framework for MLIPs in catalysis, revealing limitations in ranking accuracy for flexible or fluxional systems.

## Key findings

- UMA provides reliable rankings for 84% of rigid Ni–Cl2 complexes but drops to 53% for flexible ones.
- Ranking reliability decreases to 61% and 44% for Ru(II) and Mn(I) complexes with small energy differences.
- MLIPs show promise for rigid systems but require domain expertise for accurate interpretation in flexible systems.

## Abstract

Machine Learning
Interatomic Potentials (MLIPs) promise to transform
computational catalysis by delivering near-density functional theory
(DFT) accuracy at a fraction of the computational cost. Here, we evaluate
the Universal Machine Learning Potential for Atoms (UMA) on two data
sets of transition-metal complexes. UMA enables high-throughput evaluations
in seconds per structure on consumer-grade GPUs. Analysis of per-ligand
Spearman rank correlations (ρ > 0.6, p <
0.05) reveals variability in ranking reliability that is not captured
by aggregate metrics such as R
2 or RMSE.
However, these inaccuracies are shown to mainly occur in the near-DFT
accuracy regime where these complexes are practically indistinguishable.
For square-planar Ni complexes, reliable rankings are obtained for
84% of ligands in rigid Ni–Cl2 complexes and drop
to 53% for flexible asymmetric coordination environments, particularly
only when conformers differ by <2 kJ/mol. Data set 2 shows a similar
trend, with 61% and 44% reliability for Ru­(II) and Mn­(I) complexes,
respectively, and, as expected, challenges for fluxional systems with
small (<5 kJ/mol) relative energy gaps. These findings highlight
the promise of MLIPs for both rigid, well-defined systems and highly
flexible or fluxional catalysts, while underscoring the need to combine
the speed of ML with validation and domain expertise to ensure robust
and meaningful chemical insights.