RPSLearner: A novel approach based on random projection and deep stacking learning for categorizing NSCLC.

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Lung cancer is a leading cause of cancer death, with non-small cell lung cancer (NSCLC) being the most prevalent subtype.
• Accurate subtyping of NSCLC, particularly lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), is challenging with conventional histological and imaging methods due to limitations in definitive features and time intensity.

Purpose of the Study:

• To develop an accurate and efficient computational model for non-small cell lung cancer (NSCLC) subtyping.
• To address the diagnostic challenges posed by LUAD and LUSC using conventional methods.

Main Methods:

• Proposed RPSLearner, a novel method combining Random Projection (RP) for dimensionality reduction and stacking ensemble learning.
• Generated multiple independent RP matrices to reduce high-dimensional RNA-seq data, concatenated the resulting features, and fed them into a stack of diverse base classifiers.
• Integrated base model predictions using a deep linear layer network.

Main Results:

• RPSLearner outperformed state-of-the-art approaches in lung cancer subtype classification on a dataset of 1,333 NSCLC patients.
• Demonstrated efficient preservation of sample-to-sample distances post-dimension reduction.
• Achieved higher accuracy, F1, and AUC scores compared to individual base models and existing methods, with superior performance over conventional score ensemble techniques.

Conclusions:

• RPSLearner offers an efficient and accurate method for identifying NSCLC subtypes by integrating RP and stacking ensemble learning.
• This model shows promise for clinical diagnosis and personalized lung cancer treatment.
• The RPSLearner framework has potential for extension to the subtyping of other cancer types.