Scientists have created an innovative deep-learning system. It has been used to predict the yield of muscle stem cells differentiated from induced pluripotent stem (iPS) cells by non-invasive advanced imaging modalities. The Associate Professor Hidetoshi Sakurai is at the helm of this unique project. It is indeed the farthest step forward in stem cell research and regenerative medicine. We partnered with Epistra Inc. on the development of this AI system. Combining machine learning with nondestructive imaging, the new technique delivers remarkable efficiencies in producing differentiated stem cells.
The AI system provides extremely accurate predictions on how well it should be able to differentiate iPS cells into muscle stem cells. Understanding this mechanism is critical to many potential therapeutic uses. The system previously trained on more than 5,500 phase contrast images gathered from 34 wells throughout the differentiation period. Most importantly, it has demonstrated an incredible predictive power that allows it to correctly predict the percentage of MYF5-positive cells on day 82 of the timeline.
Advanced Imaging Techniques
The creation of this innovative AI system included an elaborate procedure of image processing and machine learning. To do so, the research team applied fast Fourier transform (FFT) to extract important morphological features. They focused their attention on analyzing phase contrast images taken between day 14 and day 38 of differentiation. This method allowed the AI to detect nuanced variations in cell morphology that are related to differentiation efficiency.
The team trained a random forest classifier on the features they extracted in order to predict muscle stem cell yield. This is a highly valuable predictive capability, since it enables researchers to flag low-efficiency samples at an early stage of the differentiation process. “Streamlining differentiation with nondestructive AI-based imaging” is a central tenet of this innovative approach, as it promises to improve overall outcomes in stem cell research.
High Predictive Accuracy
This AI system has shown an exceptional predictive accuracy, especially when looking at images taken from specific days along the differentiation timeline. As an example, images taken on day 24 turned out to be the most successful at finding the low efficiency samples. Images captured on day 31 or 34 offered a more informative window into the characteristics of high-efficiency samples.
This precision greatly increases the overall yield of high quality samples. The classifier was able to eliminate these low-quality samples with an astounding 43.7% success rate. Yet concurrently, it has boosted the production of top-quality samples by nearly three-quarters! These recent results highlight the ability of this novel AI system to revolutionize quality-control practices in the fields of regenerative medicine and cell therapy.
“Schematic overview of the experimental workflow and the machine learning system for early prediction of muscle stem cell (MuSC) differentiation efficiency.”
Future Implications
The effects of this AI prediction-based system go much deeper than just being able to make predictions. By improving the accuracy and efficiency of muscle stem cell production, researchers can potentially expedite advancements in regenerative medicine, including treatments for muscular dystrophies and other muscle-related diseases.
The collaboration between researchers and industry partners like Epistra Inc. highlights the importance of interdisciplinary efforts in pushing the boundaries of science. This technology is extremely fast moving. This would allow for the development of innovative methodologies in stem cell research, adoption of artificial intelligence to better inform outcomes.