A bibliometric analysis of artificial intelligence and machine learning in preclinical mesenchymal stem cell therapy and manufacturing from 2009 to 2025
This bibliometric analysis of 144 documents from 2009 to 2025 reveals that artificial intelligence applications in mesenchymal stem cell therapy are rapidly growing yet remain overwhelmingly preclinical, focusing on characterization rather than manufacturing quality control or clinical translation, thereby highlighting a critical gap in potency-predictive models needed to advance the field toward the clinic.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
For two decades, scientists have been working to turn a specific type of human cell into a medicine. These cells, known as mesenchymal stem cells, are found in tissues like bone marrow and fat. They hold a unique promise: they can calm the immune system and help repair damaged organs, offering hope for treating everything from joint injuries to inflammatory diseases. However, moving these cells from a laboratory dish to a patient's bedside has proven difficult. The cells are sensitive; their behavior changes depending on where they came from, how they were grown, and how many times they were divided. To be safe and effective, every batch of cells must be tested rigorously to ensure it is potent enough to work and free of defects. This testing, known as quality control, is slow, expensive, and often relies on subjective human judgment, creating a bottleneck that stops many promising therapies from reaching the clinic.
In recent years, researchers have turned to artificial intelligence and machine learning to solve this problem. These are computer systems designed to find patterns in data, much like a human would, but with the ability to process vast amounts of information quickly. The hope is that these tools could automatically analyze cells, predict how well they will work, and ensure they meet strict safety standards without the need for destructive or time-consuming tests. But while the idea is popular, no one had stopped to map exactly how this technology is actually being used in this specific field. A new study by researchers at Qatar University has done just that, creating a detailed picture of the scientific literature to see where the work is happening, what methods are being used, and where the gaps remain.
The researchers gathered and analyzed 144 scientific documents published between 2009 and 2025 that discussed the use of artificial intelligence with mesenchymal stem cells. They looked at everything from the types of computer algorithms used to the specific tasks the computers were asked to perform. The field is young but growing rapidly, with the number of publications increasing by about 35 percent each year. The work is concentrated in a few key countries, led by China and the United States, with a modest amount of collaboration between nations. Despite the excitement, the study reveals that the vast majority of this research is still happening in the laboratory, far away from the hospital bedside.
When the researchers broke down what the scientists were actually doing, a clear pattern emerged. Most of the work focused on classifying cells or diagnosing their condition based on images or data. This is a form of cell characterization, where the computer learns to recognize what a healthy cell looks like versus a damaged one. While this is useful, it is not the same as the critical step of release testing. Release testing is the final check that determines if a batch of cells is safe and potent enough to be given to a patient. The study found that very few papers addressed this specific, high-stakes task. Instead of building models to predict whether a batch would pass a regulatory release standard, researchers were mostly building models to describe the cells they were studying.
The gap between the laboratory and the clinic is stark. Nearly half of all the documents analyzed described preclinical work, meaning experiments done in labs or on animals. Only a tiny fraction, about two percent, described work that had reached the clinical stage of testing in humans. Even within the manufacturing and quality-control category, which accounted for about 13 percent of the papers, the focus remained on characterizing the cells rather than validating them for patient use. The most common sources of cells studied were bone marrow and fat tissue, and the most common applications were for bone and orthopedic issues. However, the study noted that in many cases, the specific source of the cells was not even clearly reported in the papers, making it hard to compare results or build reliable models.
The analysis suggests that while artificial intelligence is entering the world of stem cell therapy, it has not yet tackled the hardest part of the problem. The technology is being applied to the manufacturing process, but it is concentrating on describing the cells rather than solving the regulatory hurdles that prevent them from being released to patients. The researchers found that the vocabulary of artificial intelligence is now deeply woven into the conversation about stem cell biology, but it has not yet formed a strong connection with the vocabulary of manufacturing and quality control. In other words, scientists are using smart tools to understand the cells, but they have not yet built the smart tools required to certify those cells for human use.
This does not mean the technology is failing; rather, it indicates that the field is still in its early stages. The researchers point out that for artificial intelligence to truly help bridge the gap to the clinic, future work must focus on creating models that are validated against the strict criteria used to release drugs to patients. This will require better reporting of where cells come from and what their quality attributes are, so that computers can learn from consistent data. The study concludes that the path forward involves shifting attention from general cell description to the specific, regulated steps of manufacturing and release testing. Until that shift happens, the promise of artificial intelligence in this field will remain largely on the drawing board, waiting to be integrated into the systems that actually get these life-saving therapies to the people who need them.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.