GLACIER: A Multimodal Student-Teacher Foundation Model for Molecular Property Prediction

2026-06-09

AI Summary

This work addresses the limitations of existing molecular property prediction models, which often suffer from high computational costs, poor scalability, and a reliance on single-modality inputs, thereby hindering effective integration of complementary information from molecular graphs, SMILES strings, and physicochemical descriptors. To overcome these challenges, we propose the first lightweight multimodal teacher-student framework that jointly pretrains student encoders across all three modalities. Our approach incorporates a Finsler geometry-aware fusion module and a contrastive learning-based knowledge distillation strategy to efficiently transfer knowledge from large teacher models such as MiniMol and MolFormer. The resulting model achieves significant improvements in prediction accuracy and generalization while maintaining a compact architecture, demonstrating a favorable trade-off between performance and efficiency across multiple challenging molecular property prediction tasks. The code is publicly available.

Abstract

Deep learning models facilitate the discovery of molecules with tailored properties among billions of candidate compounds. However, the computational burden to develop and deploy state-of-the-art models continuously increases, limiting their scalability. Most large-scale models are unimodal in nature and overlook the potential to leverage complementary molecular data modalities. To address these shortcomings, this paper introduces the Graph-Language Alignment for Chemical Inference and Exploration using Representations (GLACIER) model, a student-teacher framework that integrates molecular graphs, SMILES strings, and physicochemical descriptors to learn rich molecular embeddings. Our framework consists of three stages: (1) we pretrain three student encoders on 100,000 drug-like molecules: a message-passing neural network for molecular graphs, a transformer-based encoder for SMILES strings, and a multilayer perceptron for physicochemical descriptors, (2) we fuse these student modalities using a novel Finsler geometry-aware module, and (3) distill complementary knowledge from large teacher models, including MiniMol and MolFormer, into a single lightweight model via contrastive learning. We demonstrate that GLACIER is a robust framework that delivers high predictive performance and computational efficiency in complex molecular property prediction tasks. Our code is publicly available at https://github.com/eemokey/glacier.

Problem

• molecular property prediction
• multimodal learning
• computational scalability
• deep learning
• molecular representation

Innovation

• multimodal fusion
• student-teacher framework
• Finsler geometry-aware module
• molecular representation learning
• knowledge distillation