A Cascade Flexible Neural Forest Model for Cancer Subtypes Classification on Gene Expression Data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate cancer subtype classification is crucial for understanding pathogenesis and tailoring treatments.
• Deep neural networks and gene expression data are increasingly used for cancer classification.
• Existing methods struggle with overfitting and low accuracy on small, high-dimensional datasets.

Purpose of the Study:

• To propose a novel model, Cascade Flexible Neural Forest (CFNForest), for improved cancer subtype classification.
• To address the limitations of existing classifiers in handling small sample sizes and high-dimensional gene expression data.
• To enhance classification accuracy and robustness in cancer subtype identification.

Main Methods:

• Developed the Cascade Flexible Neural Forest (CFNForest) model, extending flexible neural trees with ensemble strategies.
• Implemented a multilayer cascade framework to deepen the model without new hyperparameters, enabling feature transformation between levels.
• Utilized a bagging ensemble strategy and sample selection mechanisms for improved efficiency and robustness.
• Employed FNT Group Forests with diverse feature sets to enhance structural diversity for small datasets.

Main Results:

• The CFNForest model demonstrated effective improvement in cancer subtype classification accuracy on RNA-seq gene expression data.
• The proposed model showed good robustness in its classification results.
• The cascade structure and ensemble strategy contributed to enhanced performance, particularly with limited data.

Conclusions:

• CFNForest offers a robust and accurate solution for cancer subtype classification, especially for small sample, high-dimensional datasets.
• The model's architecture effectively handles complex biological data, paving the way for more precise cancer diagnostics.
• This approach holds promise for advancing cancer research and personalized medicine through improved subtype identification.