A Novel Machine Learning Based Approach For IPS Progenitor Cell Identification

Induced pluripotent stem(iPS)progenitor cells refer to induced pluripotent stem cells which have not matured at the early stage of reprogramming.The identification of iPS progenitor cells can provide valuable information for studying the origin and underlying mechanisms of iPS cells.However,due to the lack of biomarkers for progenitor cells during early reprogramming,it is currently difficult to identify the progenitor cells experimentally,and the experimental methods can only determine induced pluripotency by fluorescent probes about 6 days after reprogramming initiation.What is more,the ratio of progenitor cells during early reprograming period is below 5%,which is too low to capture experimentally in the early stage.This paper proposes a new method for identifying iPS progenitor cells based on machine learning and microscopic image analysis.Firstly,live cell imaging systems are used to record progenitor cells and other normal cells for 3 to 5 days after 48 hours of transfection with Oct4,Sox2,and Klf4 retroviruses.Then,Imaris is used to segment and track cells with calculating 11 types of cell morphology and motion features.After performing time window and feature selection,a prediction model using XGBoost is built based on the selected six types of features and best time windows.The model uses three validation methods,namely Cross-Validation,Holdout-Validation,and Independent-Test.Experimental results show that the minimum precision is higher than 52%,that is,the proportion of predicted progenitor cells within 3 to 5 days after virus infection is higher than 52%.The results also confirm that the morphology and motion pattern of progenitor cells is different from normal cells,and this difference contributes to machine learning methods to identify precursor cells.