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

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Tumors exhibit complex heterogeneous structures, necessitating advanced analytical tools.
• Single-cell RNA sequencing (scRNA-seq) provides critical insights into tumor microenvironments.
• Challenges in scRNA-seq analysis include transcriptional noise and mRNA degradation.

Purpose of the Study:

• To develop and evaluate a machine learning model for precise cell type and gene marker identification in tumors.
• To address limitations of existing scRNA-seq analysis methods.
• To compare the performance of a single layer perceptron against other advanced optimization and machine learning classifiers.

Main Methods:

• Application of machine learning classifiers, optimization algorithms, and neural networks to scRNA sequencing data.
• Development and implementation of a single layer perceptron model for gene expression profile analysis.
• Comparative analysis of the proposed single layer perceptron with models like Extra Tree Classifier, k-Nearest Neighbors, Decision Tree, Random Forest, and Gaussian Naive Bayes, each coupled with distinct optimization techniques.

Main Results:

• The proposed single layer perceptron demonstrated superior performance compared to other evaluated models.
• The model achieved high accuracy, precision, recall, and F1 scores in classifying cell types and gene markers.
• Validation was performed on normal mucosa and colorectal tumor datasets, confirming the model's efficacy.

Conclusions:

• The single layer perceptron offers a robust and accurate method for analyzing scRNA sequencing data in tumor research.
• This approach effectively overcomes challenges posed by transcriptional noise and data complexity.
• The findings support the utility of advanced machine learning for precise tumor subtyping and biomarker discovery.