Deep learning-based microarray cancer classification and ensemble gene selection approach

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Microarray data analysis faces challenges due to high dimensionality (large gene number) and limited samples.
• Accurate diagnosis and classification of genetic diseases are crucial for effective treatment.

Purpose of the Study:

• To develop a robust gene expression analysis strategy for classifying various diseases using microarray data.
• To identify key genes and classify diseases with high accuracy and generalizability.

Main Methods:

• A two-step approach involving soft ensembling for effective gene identification and a deep neural network for classification.
• Feature selection using three strategies combined with the k-nearest neighbor algorithm for gene ranking.
• Application of a stacked deep neural network for classifying multiple microarray datasets.

Main Results:

• Soft ensembling identified optimal gene subsets from diffuse large cell lymphoma, leukemia, and prostate cancer datasets.
• The deep neural network achieved high classification accuracies: 97.51% (lymphoma), 99.6% (leukemia), and 96.34% (prostate cancer).
• The model demonstrated generalizability on previously unreported datasets, including small, round blue cell tumors (SRBCTs) and multiple sclerosis-related brain lesions.

Conclusions:

• The proposed combination strategy effectively addresses microarray data challenges, enabling accurate disease classification.
• The deep neural network model exhibits strong generalizability and low error rates across diverse datasets.
• This approach offers a promising tool for diagnosing and classifying genetic malignancies and diseases.