Aim: Although new advances in diagnosis and treatment have increased, breast cancer is still an important cause of morbidity and mortality today. Proteomics, which collectively deals with relevant information about proteins, is one of the important areas of study that has been emphasized recently. It is a machine learning class that uses many layers of nonlinear processing units for deep learning, feature extraction and conversion. The aim of this study is to classify the molecular subtypes (Basal-like, human epidermal growth factor receptor 2 (HER2)-enriched, Luminal A, Luminal B) of breast cancer with the deep learning algorithm designed by using proteomic data.
Material and Method: The data set used in this study consists of published Isobaric tags for relative and absolute quantitation (iTRAQ) proteome profiling of 77 breast cancer samples by the Clinical Proteomic Tumor Analysis Consortium (NCI/NIH). The missing values in the data were completed with the mean substitution method. Lasso Regression Model was used in the selection of variables and after repeating 100 times with 10 times cross-validation method. Finally, the deep learning algorithm has been used to classify the molecular subtypes of breast cancer.
Results: The overall accuracy rate of the proposed model in classifying breast cancer are found to be 91.53%. The performance of this model for classifying molecular subtypes of breast cancer was calculated as accuracy %96.43, F-score %93.33, MCC %91.29, G-mean %93.54 for Basal-like, accuracy %94.74, F-score %84.21, MCC %81.23, G-mean %92.30 for HER2-enriched, accuracy %98.18, F-score %96.97, MCC %95.76, G-mean %98.71 for Luminal A and accuracy 93.10%, F-score 88.89%, MCC 83.89%, G-mean 91.89% for Luminal B, respectively.
Conclusion: The model designed using the deep learning algorithm has been found to perform quite well in classifying the molecular subtypes of breast cancer. In further studies, different deep learning architectures can be used to classify the molecular subtypes of breast cancer with higher accuracy.
Key words: Breast cancer; classification; deep learning; proteomics
|
