Generative AI Accelerates Medical Research Efficiency in Preterm Birth Studies

Recent investigations conducted by UC San Francisco and Wayne State University demonstrate that generative AI can analyze vast medical datasets at remarkable speeds, often surpassing the performance of traditional research teams. This innovative approach was put to the test in a study aimed at predicting preterm birth in over 1,000 pregnant women.

The researchers divided their efforts between human-only teams and those employing AI tools. Even novice researchers, including a master's student and a high school student, managed to generate effective prediction models with the assistance of AI. Within minutes, these systems produced functioning code, a task that typically consumes hours, if not days, for seasoned programmers.

The efficiency of AI lies in its capability to generate analytical code from succinct, targeted prompts. While not every AI system succeeded—only four out of eight tools yielded usable code—the ones that did operated independently, demanding little guidance from specialists.

This speed advantage enabled the junior research team to complete experiments, validate findings, and publish their results within just a few months, a timeline vastly reduced compared to traditional methods. Marina Sirota, PhD, an interim director at the Bakar Computational Health Sciences Institute (BCHSI) and co-author of the study published in Cell Reports Medicine, emphasized the relief that AI could provide from the bottlenecks faced in data science: "The speed-up couldn't come sooner for patients who need help now."

The urgency of this research is underscored by the fact that preterm birth remains the leading cause of neonatal death and contributes significantly to long-term health challenges. In the United States, about 1,000 infants are born prematurely each day, making advancements in predictive analytics crucial.

Despite ongoing research, the causes of preterm birth remain largely elusive. To identify potential risk factors, Sirota's team collated microbiome data from approximately 1,200 pregnant women sourced from nine extensive studies. Tomiko T. Oskotsky MD, who co-directs the March of Dimes Preterm Birth Data Repository, stresses that such comprehensive analysis hinges on open data sharing among researchers, combining various expertise to tackle complex challenges.

However, managing and analyzing such extensive datasets posed significant obstacles. To navigate these hurdles, the team engaged in a global challenge known as DREAM (Dialogue on Reverse Engineering Assessment and Methods), co-led by Sirota, aimed at deriving patterns specifically linked to preterm birth.

In this context, over 100 teams worldwide devised machine learning models targeting vaginal microbiome data associated with preterm birth. Although participants delivered their solutions within the three-month competition window, the aggregation of findings for publication took nearly two years.

Aiming to discover if generative AI could streamline this timeline, researchers from UCSF collaborated with Adi L. Tarca, PhD, who had spearheaded two other DREAM challenges focused on refining pregnancy stage estimation methods. Together, they tasked eight AI systems with independently generating algorithms with the same datasets used in the original DREAM challenges, without any human coding intervention.

These AI systems received meticulously crafted natural language instructions, akin to interacting with ChatGPT. The specific objectives involved analyzing the vaginal microbiome data for predictors of preterm birth and utilizing blood or placental samples for gestational age assessments. Accurate pregnancy dating is crucial, as inaccuracies can complicate care management as the pregnancy progresses.

Upon executing the AI-generated code with the DREAM datasets, it emerged that four out of the eight systems produced models that equated to, or in some cases outperformed, the results generated by human teams. Impressively, the entire process—from initial AI instruction to paper submission—was completed in just six months.

Researchers acknowledge that while AI offers considerable advantages, it does not eliminate the necessity for human oversight. These systems can potentially generate misleading outcomes, underscoring the continued need for expert validation. Nevertheless, the ability of generative AI to rapidly analyze extensive health datasets enables researchers to shift their focus from debugging code to interpreting findings and formulating significant scientific inquiries.

According to Tarca, "Thanks to generative AI, researchers with a limited background in data science won't always need to form wide collaborations or spend hours debugging code. They can focus on answering the right biomedical questions."

The authors of this study include UCSF researchers Reuben Sarwal, Claire Dubin, Sanchita Bhattacharya, MS, and Atul Butte, MD, PhD. Other contributors comprise Victor Tarca, Nikolas Kalavros, Gustavo Stolovitzky, Gaurav Bhatti, and Roberto Romero, MD, D(Med)Sc.

This research was supported by the March of Dimes Prematurity Research Center and the ImmPort program, with some data generated through the Pregnancy Research Branch of the NICHD.

Source: University of California - San Francisco. Content may be edited for clarity and length.