[2603.00857] MultiPUFFIN: A Multimodal Domain-Constrained Foundation Model for Molecular Property Prediction of Small Molecules

Computer Science > Machine Learning arXiv:2603.00857 (cs) [Submitted on 1 Mar 2026] Title:MultiPUFFIN: A Multimodal Domain-Constrained Foundation Model for Molecular Property Prediction of Small Molecules Authors:Idelfonso B. R. Nogueira, Carine M. Rebelloa, Mumin Enis Leblebici, Erick Giovani Sperandio Nascimento View a PDF of the paper titled MultiPUFFIN: A Multimodal Domain-Constrained Foundation Model for Molecular Property Prediction of Small Molecules, by Idelfonso B. R. Nogueira and 3 other authors View PDF HTML (experimental) Abstract:Predicting physicochemical properties across chemical space is vital for chemical engineering, drug discovery, and materials science. Current molecular foundation models lack thermodynamic consistency, while domain-informed approaches are limited to single properties and small datasets. We introduce MultiPUFFIN, a domain-constrained multimodal foundation model addressing both limitations simultaneously. MultiPUFFIN features: (i) an encoder fusing SMILES, graphs, and 3D geometries via gated cross-modal attention, alongside experimental condition and descriptor encoders; (ii) prediction heads embedding established correlations (e.g., Wagner, Andrade, van't Hoff, and Shomate equations) as inductive biases to ensure thermodynamic consistency; and (iii) a two-stage multi-task training this http URL prior frameworks, MultiPUFFIN predicts nine thermophysical properties simultaneously. It is trained on a multi-source dataset of 37,968 unique ...