Cancer classification of single-cell gene expression data by neural network.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression profiles are crucial for understanding cancer and guiding treatment.
• Single-cell RNA sequencing (scRNA-seq) presents challenges in gene expression analysis due to data variability.
• Machine learning approaches are increasingly used for downstream cancer analysis.

Purpose of the Study:

• To design and evaluate machine learning models for classifying 21 cancer types and normal tissues.
• To compare the performance of different machine learning algorithms, including neural networks (NN), support vector machines (SVM), k-nearest neighbors (kNN), and random forests (RF).
• To adapt the classification model for single-cell RNA sequencing data.

Main Methods:

• Developed cancer classifiers utilizing gene expression data from The Cancer Genome Atlas (TCGA).
• Trained models on 7398 cancer and 640 normal samples, focusing on the 300 most significant genes per cancer type.
• Compared NN, SVM, kNN, and RF algorithms for classification performance.
• Applied the optimized model to scRNA-seq data after kNN smoothing.

Main Results:

• The neural network (NN) model demonstrated superior and consistent performance compared to SVM, kNN, and RF.
• The developed classifier successfully identified 21 distinct cancer types and normal tissue samples.
• The approach was effectively applied to processed scRNA-seq data, maintaining high classification accuracy.

Conclusions:

• Neural network-based classification of gene expression profiles is a robust method for cancer type identification.
• The developed NN model offers a reliable tool for classifying both bulk and single-cell RNA sequencing data.
• This approach enhances cancer diagnostics and aids in understanding cancer heterogeneity.