ivysun14/project-perturb

How ivysun14/project-perturb works

This repository is a research benchmark for classifying differential-expression outcomes in single-cell perturbation data, with eight alternative methods and post-prediction calibration. The documented data workflow is DE-label generation, train/test splitting, and an optional calibration split; model-specific training and prediction scripts then produce results for analysis. The README establishes the broad benchmark and calibration relationships, while sampled source confirms selected implementation links. Many method-specific internals are unsampled, so the graph keeps those branches at their documented subsystem level.

Main entry points: calibration_split.py Calibration split · scouter_train_full.py Scouter · DE_DOC_stats.py DE label calling · train_test_split.py Train/test splits

Components 20

Data preparation

Perturbation data
DE label calling majority_vote/DE_DOC_stats.py
Train/test splits majority_vote/train_test_split.py
Calibration split majority_vote/calibration_split.py

Classification methods

Majority vote majority_vote/majority_transfer_full.py
KNN classifier KNN_vote/knn_classifier.py
Logistic regression logistic_reg/logistic_reg.py

Expression models

Scouter Scouter/scouter_train_full.py
TranScouter TranScouter/step2_train_full.py
STATE STATE/step2_train_full.py
Perturbation VAE VAE/main.py
scVI scVI/scVI_src/models/scvi/_model.py

Calibration

Sparsity calibration calibration/calibrate_sparsity.py
Variance calibration calibration/var_calibration/calibrate_var.py
DE logreg calibration calibration/logreg_calibration/calibrate_de_logreg_within.py

Analysis

Model predictions
Benchmark metrics perturb_metrics.py
Error analysis majority_vote/error_utility.py
Exploratory analysis EDA/plot_model_prediction_eda.py

Other components

Researcher

Connections 27

Researcher → DE label calling: runs README.md
Perturbation data → DE label calling: tests for DE README.md
DE label calling → Train/test splits: labels perturbations README.md
Train/test splits → Calibration split: carves calibration set README.md
Calibration split → Majority vote: supplies split data README.md
Calibration split → KNN classifier: supplies split data README.md
Calibration split → Logistic regression: supplies split data README.md
Calibration split → Scouter: supplies split data README.md
Calibration split → TranScouter: supplies split data README.md
Calibration split → STATE: supplies split data README.md
Calibration split → Perturbation VAE: supplies split data README.md
Calibration split → scVI: supplies split data README.md
Majority vote → Model predictions: produces README.md
KNN classifier → Model predictions: produces README.md
Logistic regression → Model predictions: produces README.md
Scouter → Model predictions: produces README.md
TranScouter → Model predictions: produces README.md
STATE → Model predictions: produces README.md
Perturbation VAE → Model predictions: produces README.md
scVI → Model predictions: produces README.md
Model predictions → Sparsity calibration: calibrates zeros calibration/calibrate_sparsity.py
Model predictions → Variance calibration: calibrates variance README.md
Model predictions → DE logreg calibration: calibrates DE README.md
Model predictions → Benchmark metrics: evaluated by perturb_metrics.py
Model predictions → Error analysis: analyzed for errors majority_vote/error_utility.py
Model predictions → Exploratory analysis: visualized by EDA/plot_model_prediction_eda.py
Scouter → Benchmark metrics: metric calls

More Python diagrams: jgyates/genmon · patrick-kidger/optimistix · sentient-engineering/sentient · tencent-hunyuan/hunyuan3d-omni · zerosumquant/claude-conversation-extractor

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·