How xieresearchgroup/code-ae works
CODE-AE is a research implementation for learning context-aware gene-expression representations and predicting personalized drug response. Its principal workflow is to prepare expression and response data, pre-train an encoder across contexts, fine-tune a response predictor for selected drugs, and produce predictions or benchmark results. The README also documents TCGA inference, encoded-feature export, ADAE experiments, and baseline comparisons. Source excerpts show model and data-loader behavior; unsampled modules are included where documented, but their internal calls are not inferred.
Main entry points: pretrain_hyper_main.py Pretraining entrypoint · dsn_ae.py Domain-separation AE · data.py Response data loaders · drug_ft_hyper_main.py Fine-tuning entrypoint
Components 29
Data preparation
- Dataset configuration
code/data_config.py - Expression preprocessing
code/data_preprocessing.py - Response data loaders
code/data.py
Representation learning
- Pretraining entrypoint
code/pretrain_hyper_main.py - AE and VAE models
code/ae.py - Domain-separation AE
code/dsn_ae.py - AE training
code/train_ae.py - ADAE training
code/train_adae.py - CODE base training
code/train_code_base.py - CODE adversarial training
code/train_code_adv.py - CODE MMD training
code/train_code_mmd.py - CORAL training
code/train_coral.py - Denoising AE training
code/train_dae.py - DSN training
code/train_dsn.py - DSNA training
code/train_dsna.py - VAE training
code/train_vae.py
Drug prediction
- Fine-tuning entrypoint
code/drug_ft_hyper_main.py - Response fine-tuning
code/fine_tuning.py - TCGA inference entrypoint
code/drug_inference_main.py - Prediction inference
code/inference.py
Evaluation and outputs
- Encoded feature export
code/generate_encoded_features.py - ML baselines
code/ml_baseline.py - MLP benchmark
code/mlp_main.py - TCRP benchmark
code/tcrp_main.py - VAEN benchmark
code/vaen_main.py - Evaluation metrics
code/evaluation_utils.py
Other components
- Researcher
- Benchmark data
- Predictions and results
Connections 36
- Researcher → Pretraining entrypoint: runs pretraining
README.md - Pretraining entrypoint → Response data loaders: prepares data
code/pretrain_hyper_main.py - Pretraining entrypoint → AE training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → ADAE training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → CODE base training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → CODE adversarial training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → CODE MMD training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → CORAL training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → Denoising AE training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → DSN training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → DSNA training: selects method
code/pretrain_hyper_main.py - Pretraining entrypoint → VAE training: selects method
code/pretrain_hyper_main.py - AE training → AE and VAE models: trains model
code/ae.py - ADAE training → AE and VAE models: trains model
code/train_adae.py - CODE base training → Domain-separation AE: trains model
code/train_code_base.py - CODE adversarial training → Domain-separation AE: trains model
code/train_code_adv.py - CODE MMD training → Domain-separation AE: trains model
code/train_code_mmd.py - DSN training → Domain-separation AE: trains model
code/train_dsn.py - DSNA training → Domain-separation AE: trains model
code/train_dsna.py - VAE training → AE and VAE models: trains model
code/train_vae.py - Researcher → Fine-tuning entrypoint: runs fine-tuning
README.md - Fine-tuning entrypoint → Response data loaders: prepares data
code/drug_ft_hyper_main.py - Fine-tuning entrypoint → Response fine-tuning: fine-tunes encoder
README.md - Response fine-tuning → Evaluation metrics: evaluates model
code/fine_tuning.py - Researcher → TCGA inference entrypoint: runs inference
README.md - TCGA inference entrypoint → Prediction inference: generates predictions
- TCGA inference entrypoint → Predictions and results: produces predictions
README.md - Researcher → Encoded feature export: exports features
README.md - Encoded feature export → Predictions and results: writes features
README.md - Researcher → ML baselines: runs baseline
README.md - Researcher → MLP benchmark: runs benchmark
README.md - Researcher → TCRP benchmark: runs benchmark
README.md - Researcher → VAEN benchmark: runs benchmark
README.md - Benchmark data → Response data loaders: provides datasets
- Dataset configuration → Expression preprocessing: configures sources
code/data_preprocessing.py - Dataset configuration → Response data loaders: configures sources
code/data.py