LungDWM

LungDWM integrates weakly paired multiomics data to improve lung cancer subtype diagnosis by extracting omics-specific and shared features, imputing missing omics views, and fusing resulting representations.

Key Features:

• Attention-Based Feature Encoders: Employs attention-based encoders tailored to each omics modality to identify diagnostic features and extract shared and complementary information across omics layers.
• Generative Adversarial Learning for Data Imputation: Uses generative adversarial learning (GANs) to impute missing omics data from available views using extracted features.
• Individual Loss Function: Incorporates an individual loss function to preserve omics-specific characteristics during feature learning.
• Fusion of Extracted and Imputed Features: Fuses extracted and imputed features to produce integrated representations for lung cancer subtype classification.

Scientific Applications:

• Precision Oncology: Supports molecular stratification of lung cancer subtypes to inform treatment decisions and personalized therapy selection.
• Multiomics Integration with Incomplete Data: Enables diagnostic analysis and subtype classification using incomplete, weakly paired multiomics datasets common in clinical settings.

Methodology:

Applies attention mechanisms and attention-based encoders, generative adversarial networks (GANs) for imputation, an individual loss function to retain omics-specific information, and fusion of extracted and imputed features to handle weakly paired multiomics data.