MolFM: A Multimodal Molecular Foundation Model

6 Jun 2023  ·  Yizhen Luo, Kai Yang, Massimo Hong, Xing Yi Liu, Zaiqing Nie ·

Molecular knowledge resides within three different modalities of information sources: molecular structures, biomedical documents, and knowledge bases. Effective incorporation of molecular knowledge from these modalities holds paramount significance in facilitating biomedical research. However, existing multimodal molecular foundation models exhibit limitations in capturing intricate connections between molecular structures and texts, and more importantly, none of them attempt to leverage a wealth of molecular expertise derived from knowledge graphs. In this study, we introduce MolFM, a multimodal molecular foundation model designed to facilitate joint representation learning from molecular structures, biomedical texts, and knowledge graphs. We propose cross-modal attention between atoms of molecular structures, neighbors of molecule entities and semantically related texts to facilitate cross-modal comprehension. We provide theoretical analysis that our cross-modal pre-training captures local and global molecular knowledge by minimizing the distance in the feature space between different modalities of the same molecule, as well as molecules sharing similar structures or functions. MolFM achieves state-of-the-art performance on various downstream tasks. On cross-modal retrieval, MolFM outperforms existing models with 12.13% and 5.04% absolute gains under the zero-shot and fine-tuning settings, respectively. Furthermore, qualitative analysis showcases MolFM's implicit ability to provide grounding from molecular substructures and knowledge graphs. Code and models are available on https://github.com/BioFM/OpenBioMed.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Text-based de novo Molecule Generation ChEBI-20 MolFM-Base Text2Mol 58.3 # 4
BLEU 82.2 # 8
Exact Match 21.0 # 11
Levenshtein 19.445 # 8
MACCS FTS 85.4 # 11
RDK FTS 69.7 # 14
Morgan FTS 75.8 # 2
Validity 89.2 # 10
Parameter Count 296200000 # 15
Molecule Captioning ChEBI-20 MolFM-Base BLEU-2 58.5 # 10
BLEU-4 49.8 # 10
ROUGE-1 65.3 # 8
ROUGE-2 50.8 # 9
ROUGE-L 59.4 # 7
METEOR 60.7 # 10
Text2Mol 57.6 # 6
Molecule Captioning ChEBI-20 MolFM-Small BLEU-2 54.2 # 16
BLEU-4 45.2 # 17
ROUGE-1 62.3 # 14
ROUGE-2 46.9 # 15
ROUGE-L 56.2 # 15
METEOR 56.4 # 17
Text2Mol 55.7 # 9
Text-based de novo Molecule Generation ChEBI-20 MolFM-Small Text2Mol 57.3 # 9
BLEU 80.3 # 11
Exact Match 16.9 # 13
Levenshtein 20.868 # 7
MACCS FTS 83.4 # 14
RDK FTS 66.2 # 15
Morgan FTS 72.1 # 8
Validity 85.9 # 13
Parameter Count 13620000 # 4

Methods