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()andn_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},
}