Welcome to Ageas’s documentation!

Ageas (AutoML-based Genetic regulatory Element extrAction System) is a computational framework for inferring cell fate bias from static single-cell and spatial multi-omics data.

Ageas trains a heterogeneous panel of classifiers — neural networks (MLP/RNN, ResNet-mixer) alongside classical models (XGBoost, Logistic Regression, SVM, Naive Bayes) — over an Hangar of candidate units. A Deck orchestrates the sortie pipeline of n-iteration k-fold cross-validation, retains the strongest units, and aggregates per-class explanations (Integrated Gradients, SHAP, or model coefficients) into a unified factor importance table.

Note

This project is under active development.

Key Features

  • Heterogeneous Model Panel: Mixes deep learning classifiers (NN_Classifier, Mixer_Classifier) with classical estimators (XGB_Classifier, LogReg_Classifier, SVM_Classifier, MNB_Classifier) under a single Lightning-style API.

  • N-Iteration K-Fold Selection: n_kfold_selection() runs successive cross-validation rounds and prunes the squad to the units that survive the configured retention/cutoff thresholds.

  • Iterative Factor Extraction: n_iter_extraction() and n_iter_boost_selection() chain selection and explanation passes to converge on the most informative regulatory factors per cell class.

  • Unified Explanation Pipeline: debrief() weights per-unit explanations by their validation/test metrics and produces a single integrated importance table across the squad.

Citation

If you use Ageas in your research, please cite:

@software{ageas,
  title={Ageas enables time-agnostic cell fate inference from single-cell and spatial multi-omics data},
  author={Junyao Jiang, Alex Kong, Jack Yu},
  url={https://github.com/MaftyLab/Ageas},
}

Contents