DCGN

DCGN classifies cancer subtypes from high-dimensional gene expression data to enable subtype identification and downstream biological interpretation.

Key Features:

• Nonlinear Dimensionality Reduction: DCGN integrates convolutional neural networks (CNNs) and bidirectional gated recurrent units (BiGRUs) to perform nonlinear dimensionality reduction of high-dimensional gene expression data.
• Feature Extraction and Retention: The CNN component extracts local features from gene expression profiles while the BiGRU processes these deep features to retain sequential and contextual information relevant to subtype identification.
• Handling Sparse Data: A synthetic minority oversampling technique (SMOTE) preprocessing step equalizes class distributions to mitigate issues from sparse datasets and small sample sizes.
• Comprehensive Evaluation: Performance was compared against seven other cancer subtype classification methods using breast and bladder cancer gene expression datasets, with DCGN demonstrating superior classification results.

Scientific Applications:

• Aetiology and Tumour Biology: Identifies distinct cancer subtypes to inform studies of cancer origins and underlying biological mechanisms.
• Prognosis: Provides subtype assignments that can be used to assess disease progression and correlations with patient outcomes.
• Personalized Treatment: Enables patient stratification by subtype to support development of tailored therapeutic strategies.

Methodology:

Convolutional neural networks (CNNs), bidirectional gated recurrent units (BiGRUs), and the synthetic minority oversampling technique (SMOTE) were applied; performance was evaluated against seven other classifiers using breast and bladder cancer gene expression datasets.