A new multi-omics AI model combines cfDNA and cfRNA to predict preterm birth risk with nearly 90% accuracy, offering promising insights for maternal health
This new AI model, with its ability to predict preterm birth with nearly 90% accuracy, represents a significant leap forward in maternal health care. By analysing both cell-free DNA (cfDNA) and cell-free RNA (cfRNA), the model provides a deeper understanding of the risk factors in each pregnancy. The early identification of high-risk pregnancies could lead to earlier interventions, potentially improving outcomes for both mothers and babies and reducing complications associated with preterm birth.
The research was published in npj Digital Medicine.
Preterm birth affects 15 million babies worldwide
Preterm birth is a leading cause of maternal and neonatal morbidity and mortality worldwide. Each year, around 15 million babies are born prematurely, accounting for roughly 11% of all births worldwide, according to a review study. The earlier a baby is born, the greater the health risks.
Detecting high-risk pregnancies as early as possible is crucial. Now, the combination of large language models and multi-omics data could open new pathways for predicting disease risk and improving maternal and fetal health outcomes. Predicting preterm birth remains difficult due to its complexity. No single marker has been sufficient to determine risk accurately.
Combining multi-omics and AI
Researchers have come together to develop GeneLLM, a specialised large language model that interprets complex biological data to identify pregnancies at risk of preterm birth (PTB). By examining genetic material in the mother’s bloodstream—cell-free DNA (cfDNA) and cell-free RNA (cfRNA), the team was able to create predictive models with remarkable accuracy.
The study, which followed a nested case-control design, included 682 pregnant women and involved sequencing plasma samples for cfDNA and cfRNA. Three separate models were constructed: one using cfDNA alone, another using cfRNA alone, and a third integrating both cfDNA and cfRNA.
All three models utilised a Transformer-based architecture and delivered strong predictive performance, each exceeding 80% accuracy. Specifically, the cfDNA-only model achieved an AUC of 0.822, the cfRNA-only model reached 0.851, while the combined cfDNA + cfRNA model attained the highest AUC of 0.89. (An AUC closer to 1.0 indicates greater precision and reliability, making these results highly promising for the model’s clinical application.
The findings highlight that combining cfDNA and cfRNA offers complementary insights, producing the most accurate predictions of preterm birth risk and demonstrating the potential of multi-omics AI approaches in maternal healthcare.
The researchers found that RNA editing levels were markedly higher in preterm cases, and models based on RNA editing features achieved an AUC of 0.82, outperforming single-omics models. These findings underscore the transformative potential of AI and multi-omics integration in prenatal medicine, boosting preterm birth prediction to nearly 90%.
Dr. Zhou Si, Chief Scientist at BGI Genomics’ IIMR and first author of the study, explained that “Our study shows that integrating cfDNA and cfRNA with LLM outperforms conventional methods in predicting PTB. Importantly, the model is efficient, resource-light, and ready for clinical translation. Beyond prediction, our findings also reveal RNA editing as a promising new target for understanding and regulating PTB.”
