HetEnc

HetEnc applies deep learning to integrate multi-platform gene expression datasets by separating information domains and enabling single-platform prediction.

Key Features:

• Integrated Multi-Platform Data Analysis: Integrates multi-platform gene expression datasets to produce unified representations for downstream analysis.
• Unsupervised Feature Representation Module: Constructs three distinct encoding networks via unsupervised learning to transform raw gene expression data into high-level abstracted features.
• Supervised Neural Network Module: Trains a six-layer fully-connected feed-forward neural network on the abstracted features to predict biological endpoints.
• Domain Separation and Feature Abstraction: Performs information-domain separation and feature abstraction from multi-platform data to isolate relevant representations for prediction.
• Single-Platform Prediction Capability: After multi-platform training, enables predictions using data from a single platform.

Scientific Applications:

• Integrated gene expression analysis: Facilitates cross-platform integration and prediction in gene expression studies.
• Benchmarking on SEQC neuroblastoma dataset: Validated on the SEQC neuroblastoma dataset and shown to outperform traditional machine learning approaches.

Methodology:

Two-stage approach using unsupervised feature abstraction that constructs three encoding networks to generate high-level representations from multi-platform gene expression data, followed by supervised training of a six-layer fully-connected feed-forward neural network on those abstracted features to predict biological endpoints, with domain separation enabling single-platform prediction.