Scientists at the University of Virginia have developed a cutting-edge machine learning approach to identify heart drugs that can minimize harmful scarring after a heart attack or other injuries. This groundbreaking approach has already identified a promising candidate that could prevent harmful heart scarring in a unique way compared to previous drugs. Their findings, published in the journal Proceedings of the National Academy of Sciences, highlight the potential of this new tool to predict and explain the effects of drugs for various diseases.
Heart disease, metabolic disease, and cancer are complex and challenging to treat. By employing machine learning, researchers can reduce this complexity and identify essential factors contributing to the diseases while gaining a better understanding of how drugs can modify diseased cells. The collaboration of machine learning with human learning not only helps to identify drugs but also explains how they work. This knowledge is crucial for designing clinical trials and identifying potential side effects.
In this study, the team combined a computer model based on decades of human knowledge with machine learning to better understand how drugs affect fibroblasts, a type of cell that repairs the heart after injury. While fibroblasts aid in the healing process, they can also cause harmful scarring called fibrosis. Previous attempts to identify drugs targeting fibroblasts have focused on specific aspects of their behavior without a clear understanding of how these drugs work.
To address this knowledge gap, the researchers developed a new approach called logic-based mechanistic machine learning. They examined the effects of thirteen promising drugs on human fibroblasts and trained the machine learning model with this data to predict how the drugs influence the cells and their behavior. By utilizing this model, the researchers discovered new insights into how the drug pirfenidone, already approved by the FDA for pulmonary fibrosis, suppresses contractile fibers within fibroblasts, ultimately stiffening the heart. Additionally, they predicted how another type of contractile fiber could be targeted by the experimental inhibitor WH4023, which they validated through experiments with human cardiac fibroblasts.
While further research is needed to verify these drugs’ efficacy in animal models and human patients, this study demonstrates the potential of mechanistic machine learning to discover cause-and-effect relationships in biology. It highlights the significant advancements this technology can bring not only to heart injury treatments but also to the development of new therapies for various diseases.
The researchers are optimistic about the future possibilities and plan to test whether pirfenidone and WH4023 can also suppress fibroblast contraction in scarred tissue using preclinical animal models. This research serves as an excellent example of how machine learning and human learning can collaborate to not only discover new drugs but also comprehend how they function. It paves the way for improved treatment strategies and a better understanding of complex diseases like heart fibrosis.
