OmniCellTOSG: The First Cell Text-Omic Signaling Graphs Dataset for Graph Language Foundation Modeling.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Large-scale single-cell omic datasets are rapidly growing, driving the need for advanced analytical tools.
• Existing omic foundation models (FMs) primarily use numerical transcriptomic data, often neglecting crucial biomedical prior knowledge and signaling interactions.
• Integrating diverse biological data types is essential for deeper scientific discovery and precision medicine.

Purpose of the Study:

• To introduce a novel data structure, Text-Omic Signaling Graph (TOSG), for unifying biomedical text, omic data, and signaling networks.
• To construct a large-scale resource, OmniCellTOSG, using the TOSG framework.
• To develop a multimodal graph language FM, CellTOSG-FM, for analyzing integrated biological data.

Main Methods:

• Developed the Text-Omic Signaling Graph (TOSG) data structure.
• Created OmniCellTOSG, a resource of ~0.5 million meta-cell TOSGs from ~80 million single-cell RNA-seq profiles.
• Built CellTOSG-FM, a multimodal graph language foundation model.

Main Results:

• CellTOSG-FM demonstrated superior performance compared to existing omic FMs on various downstream tasks.
• The model successfully integrated textual, omic, and signaling network information.
• Achieved interpretable insights into disease-associated targets and signaling pathways.

Conclusions:

• The TOSG framework and CellTOSG-FM offer a powerful approach for analyzing complex biological data.
• This multimodal integration advances research in life sciences and precision medicine.
• Provides interpretable biological insights crucial for drug discovery and disease understanding.