QuantitativeBiology/GainPro

DANN & GAIN Hybrid Model

This repository contains the GAIN-DANN hybrid model, which is a hybrid model that combines:

• Generative Adversarial Imputation Networks (GAIN) for missing values imputation

• Domain-Adversarial Neural Networks (DANN) for domain-aware learning and classification.

The model is designed for proteomics data and has been pre-trained on a subset of the HeLa dataset.

---

Model Architecture

Key components:

• Encoder: Maps input to latent space
• Generator & Discriminator: For adversarial imputation
• Gradient Reversal Layer: Enables domain adversarial training
• Domain Classifier: Predicts domain labels
• Decoder: Reconstructs inputs after imputation

---

Model Configuration

The pre-trained model (pytorch_model.bin) has been trained on the HeLa dataset with the following configuration:

• Input Dimension: 3013
• Latent Dimension: 3013
• Number of Classes: 17
• Hidden Dimension: 128
• Dropout: 0.3

The model weights are stored in pytorch_model.bin, and the configuration is provided in config.json.

---

Usage

To use the pre-trained model for inference:

1. Install the required dependencies:

   pip install torch numpy