A Comparison of Supervised Classification Methods for Classifying Biological Cell Types

Abstract

The advent of single-cell RNA sequencing has enabled researchers to characterize specific cell types and learn more about their functional and pathological roles. Therefore, correctly classifying these cells into cell types is essential. Past research has focused on developing robust methods to classify cell types, hierarchies, and annotations, focusing on generalizability. In this article, different reduction methods and classifiers on a scRNA-seq dataset are evaluated, and the classifiers are k-nearest neighbors (KNN), neural network (NN), AdaBoost, and support vector machines (SVM). Optimizing SVM with cross-validation will also be discussed. Results showed that a Principal Component Analysis-reduced dataset and a Support Vector Machine (SVM) with a linear kernel performed better than others. This article offers potential directions for selecting classifiers and reduction methods to work with biological data and additional analytical insights.