A novel sequence-based transformer model architecture for integrating multi-omics data in preterm birth risk...

Area of Science:

• Biomedicine
• Genomics
• Artificial Intelligence

Background:

• Preterm birth (PTB) is a major cause of infant mortality and morbidity.
• Current PTB prediction methods have limitations.
• Large language models (LLMs) show promise for disease risk prediction.

Purpose of the Study:

• To develop and evaluate a novel transformer-based LLM for PTB risk prediction using multi-omics data.
• To assess the performance of integrating cell-free DNA (cfDNA) and cell-free RNA (cfRNA) for PTB prediction.
• To explore the utility of RNA editing in cfDNA and cfRNA integration for PTB risk assessment.

Main Methods:

• Developed a transformer-based architecture for integrating cfDNA and cfRNA sequencing data.
• Trained and tested LLM models on cfDNA, cfRNA, and integrated multi-omics data.
• Evaluated model performance using Area Under the Curve (AUC).

Main Results:

• The cfDNA LLM achieved an AUC of 0.822; the cfRNA LLM achieved an AUC of 0.851.
• Integrated cfDNA and cfRNA data achieved a superior AUC of 0.890.
• Integration using RNA editing yielded an AUC of 0.82.

Conclusions:

• Transformer-based LLMs effectively integrate multi-omics data for PTB risk prediction.
• Multi-omics data fusion significantly enhances PTB prediction accuracy.
• AI-driven multi-omics approaches hold potential for precision obstetrics and biomedicine.