A stacking ensemble deep learning approach to cancer type classification based on TCGA data.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Morphological characteristics alone present limitations for accurate cancer tumor classification.
• Accurate cancer type classification is crucial for effective cancer diagnosis and therapeutic strategies.
• Five common cancers in women include breast, lung, colorectal, thyroid, and ovarian cancers.

Purpose of the Study:

• To develop and evaluate a stacking ensemble deep learning model for multi-class classification of five common female cancers using RNASeq data.
• To compare the performance of the proposed model against single 1D-CNN and various machine learning methods.

Main Methods:

• Utilized RNASeq gene expression data from the Pan-Cancer Atlas via the TCGAbiolinks package in R.
• Implemented a stacking ensemble deep learning model based on one-dimensional convolutional neural network (1D-CNN).
• Employed the Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection and compared models with and without LASSO.

Main Results:

• The proposed stacking ensemble model demonstrated superior performance in classifying the five common female cancers compared to single 1D-CNN and traditional machine learning classifiers.
• Machine learning methods, particularly Support Vector Machines with Radial Basis Function (SVM-R) and Artificial Neural Networks (ANN), showed better performance with under-sampling techniques than over-sampling.
• Statistical significance tests confirmed notable accuracy differences between various machine learning models, highlighting the effectiveness of certain approaches.

Conclusions:

• The developed stacking ensemble deep learning model offers a promising approach for accurate multi-class cancer classification using RNASeq data.
• The findings suggest that this model can significantly aid in the early detection and diagnosis of common cancers in women.
• Improved diagnostic capabilities can lead to the design of more effective early treatment strategies, potentially enhancing patient survival rates.