Prediction of Bone Formation Rate of Artificial Bone With Machine Learning Models Considering the Variation of Experimental Results

TL;DR

This study uses machine learning to predict the variation in bone formation rates of artificial bone materials, aiming to improve their development and reduce reliance on animal testing.

Contribution

The study introduces the use of Jensen–Shannon divergence as a metric to evaluate model accuracy for predicting variation in bone formation rates.

Findings

A machine learning model was developed to predict variation in bone formation rates using multiple sub-datasets.

Jensen–Shannon divergence was validated as an effective metric for evaluating model accuracy.

An optimal combination of explanatory variables was identified to enhance predictive accuracy.

Abstract

The proportion of older people in the world's total population is expected to increase. Bone diseases are more prevalent in older people; therefore, the number of patients with such diseases is expected to increase worldwide. Artificial bone is a biomaterial used in the treatment of bone diseases. Artificial bones with high bone formation rates are desired; however, the results of artificial bone implantation vary. There are also ethical issues associated with animal experiments. Our purpose in this study is to predict the variation in bone formation rates. We created multiple sub‐datasets and constructed a machine learning model to predict the variation in bone formation rates by considering the results of multiple measurements. We also propose a metric, Jensen–Shannon (JS) divergence, to evaluate the accuracy of the model for predicting variation. We tested the validity of JS divergence by comparing combinations of explanatory variables. Additionally, we found an optimal combination of explanatory variables to construct a model with high predictive accuracy. We expect that the prediction of variation will be useful for improving the practical development of materials and medicines, such as artificial bones, for which stable effects are required, regardless of the individual.

Figures