Machine Learning Accelerates Screening of Osteoclast Differentiation Inhibitors from Natural Products

Natural products that inhibit osteoclast differentiation are promising therapeutic and preventive agents for osteoporosis. Conventionally, identifying osteoclast differentiation involves visual inspection of the microscope images of stained osteoclasts. In this study, a supervised machine learning model was developed to classify bright-field microscope images of osteoclasts without staining. The model was used to screen a compound library, and osteoclast differentiation inhibitors were identified, demonstrating the validity of our method. Next, an in-house library of fungal extracts was screened, and pinolidoxin was revealed as an inhibitor of osteoclast differentiation. Our machine learning method enabled accurate, objective, and high-throughput evaluation of osteoclast differentiation and efficient screening of the inhibitors from natural product extracts. This study represents the first machine learning classification developed to evaluate the inhibitory activity of natural products in osteoclast differentiation.