jruffle/ae-tracerx-256d

Autoencoder (TRACERx-focused, 256D)

This model is part of the TRACERx Datathon 2025 transcriptomics analysis pipeline.

Model Details

• Model Type: Autoencoder
• Dataset: TRACERx-focused
• Latent Dimensions: 256
• Compression Mode: transcriptome
• Framework: PyTorch

Usage

This model is designed to be used with the TRACERx Datathon 2025 analysis pipeline. It will be automatically downloaded and cached when needed.

Model Architecture

• Input: Gene expression data
• Hidden layers: [input_size, 512, 256, 128, 256]
• Output: 256-dimensional latent representation
• Activation: ELU with batch normalization

Training Data

Trained exclusively on TRACERx open dataset

Files

• autoencoder_256_latent_dims_oos_mode.pt: Main model weights
• latent_df.csv: Example latent representations (if available)