Tumor cell type and gene marker identification by single layer perceptron neural network on single-cell RNA sequence data

Abstract

Tumors have drawn increasing attention recently because of their heterogeneous interior structures. Particularly, single-cell RNA (scRNA) mechanics have made important contributions to the field of tumor research. To investigate the cell types and identify similar types of gene markers present inside a tumor, machine learning classifier, optimization, and neural network models were applied to scRNA sequencing data. Indeed, even though single-cell analysis is a more powerful tool, several issues have been identified, such as transcriptional noise that alters gene expression and degrades mRNA. Recently, optimization models for single-cell analysis have been developed to address these kinds of issues, and encouraging results have been reported. scRNA sequencing is popular because it produces biological information in the form of patterns that are displayed within the transcriptome profile. The neural network approach plays an important role in understanding and identifying these distinct patterns. A single layer perceptron was introduced to better analyze the data pattern within gene expression profiles. Finally, recently developed optimization models with machine learning classifiers are compared with the proposed single layer perceptron. The single layer perceptron performs better compared with other models such as extra tree classifier with genetic algorithm, k-nearest neighbors with bat optimization, decision tree with gray wolf optimization, random forest with firefly optimization, and Gaussian naïve Bayes with artificial bee colony optimization. This study also focused on classifying these unique cell types and gene markers using scRNA sequence datasets. The proposed single layer perceptron was assessed using two datasets: normal mucosa and colorectal tumors. Our findings showed that the proposed single layer perceptron performed exceptionally well with accuracy, precision, recall, and F1 value.