deeptan-network 0.1.1

DeepTAN: A multi-task framework designed to infer multi-omics trait-associated networks and reconstruct phenotype-specific omics states from mosaic data input.

DeepTAN

Describe

DeepTAN is a graph-based framework that learns gene representations and reconstructs trait-aware gene networks from omics data. Input features are filtered and projected to generate a guidance graph and quantitative embeddings. Stacked GATv2Conv layers refine embeddings via multi-head attention, producing biological state–specific representations. An adaptive multi-scale subgraph pooling strategy captures hierarchical network structure by aggregating local subgraphs into global graph embeddings. The learned representations are jointly optimized for phenotype prediction, sample clustering, and feature imputation through a multi-task learning with dynamic balancing strategy. DeepTAN generates biological-specific gene interactions in which edge weights quantitatively reflect interaction strength, enabling downstream trait-aware network inference.

Installation

conda create -n deeptan python=3.13 -y
conda activate deeptan

pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cu128
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.7.0+cu128.html
pip install deeptan-network

Usage

Please checkout the documentations at https://github.com/wangying608/deeptan

Asking for help

If you have any questions, please contact us via GitHub issues or email us.