Skip to content

Latest commit

 

History

History
86 lines (66 loc) · 3.32 KB

README.md

File metadata and controls

86 lines (66 loc) · 3.32 KB
Raw

CODE-AE

Shield: CC BY-NC 4.0

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

CC BY-NC 4.0

CONTEXT-AWARE DECONFOUNDING AUTOENCODER

@article{he2022context,
  title={A context-aware deconfounding autoencoder for robust prediction of personalized clinical drug response from cell-line compound screening},
  author={He, Di and Liu, Qiao and Wu, You and Xie, Lei},
  journal={Nature Machine Intelligence},
  volume={4},
  number={10},
  pages={879--892},
  year={2022},
  publisher={Nature Publishing Group UK London}
}

1. Introduction

CODE-AE is a Python implementation of Context-aware Deconfounding Autoencoder (CODE-AE) that can extract common biological signals masked by context-specific patterns and confounding factors.

Architecture

architecture

3. Installation

CODE-AE depends on Numpy, SciPy, PyTorch (CUDA toolkit if use GPU), scikit-learn, pandas. You must have them installed before using CODE-AE.

The simple way to install them is using Docker: the Dockerfile is given within folder code/

4. Usage

4.1. Data

Benchmark datasets available at Zenodo[http://doi.org/10.5281/zenodo.4776448] (version 2.0)

4.2 Basic Usage

4.2.1 Pre-train encoders

    $ python pretrain_hyper_main.py

Arguments in this script:

  • --method: method to be used for encoder pre-training, choose from ['code_adv', 'dsna', 'dsn', 'code_base', 'code_mmd', 'adae', 'coral', 'dae', 'vae','vaen', 'ae']
  • --train: retrain the encoders
  • --no-train: no re-training
  • --norm: use L2 normalization for embedding
  • --no-norm: don't use L2 normalization for embedding

4.2.1 Fine-tune encoders for different drug response predictive models

    $ python drug_ft_hyper_main.py

Arguments in this script:

  • --method: method to be used for encoder pre-training, choose from ['code_adv', 'dsna', 'dsn', 'code_base', 'code_mmd', 'adae', 'coral', 'dae', 'vae','vaen', 'ae']
  • --train: retrain the encoders
  • --no-train: no re-training
  • --norm: use L2 normalization for embedding
  • --no-norm: don't use L2 normalization for embedding
  • --metric: metric used in early stopping
  • --pdtc: fine-tune PDTC drugs (target therapy agents)
  • --no-pdtc: fine-tune chemotherapy drugs

Additional scripts

  • adae_hyper_main.py: run ADAE de-confounding experiments
  • drug_inference_main.py: generate TCGA sample prediction
  • generate_encoded_features.py: generate encoded features by pre-trained encoders
  • ml_baseline.py: get benchmark experiments' results using random forest and elastic net
  • mlp_main.py: get benchmark experiments' results using simple multi-layer neural network
  • tcrp_main.py: get benchmark experiments' results using TCRP
  • vaen_main.py: get benchmark experiments' results using VAEN (VAE+elastic net)