Supervised learning to predict sperm sorting by magnetophoresis

Abstract

Machine learning is gaining popularity in the commercial world, but its benefits are yet to be well-utilised by many in the microfluidics community. There is immense potential in bridging the gap between applied engineering and artificial intelligence as well as statistics. We illustrate this by a case study investigating the sorting of sperm cells for assisted reproduction. Slender body theory (SBT) is applied to compute the behavior of sperm subjected to magnetophoresis, with due consideration given to statistical variations. By performing computations on a small subset of the generated data, we train an ensemble of four supervised learning algorithms and use it to make predictions on the velocity of each sperm. Our results suggest that magnetophoresis can magnify the difference between normal and abnormal cells, such that a sorted sample has over twice the proportion of desirable cells. In addition, we demonstrated that the predictions from machine learning gave comparable results with significantly lower computational costs.