AURORA: AI Unification and Reconstruction of Omics Reassembly Atlas

Package: AURORA

AURORA is a Python package for multimodal omics integration and cross-modality reconstruction.

Requirements

• Linux/UNIX/Windows system
• Python >= 3.8
• torch

Create environment

conda create -n AURORA python
conda activate AURORA

Installation

Install from the project root:

pip install .

Alternatively, you can also install the package directly from GitHub:

pip install git+https://github.com/JackieHanLab/Aurora.git

Environment has been tested

AURORA.yml

Usage

Step 1: Configure data

import aurora

modalityA_h5ad.layers['input'] = modalityA_h5ad.X
aurora.configure_dataset(modalityA_h5ad, use_layers = 'input', use_uid = <sample_ID>,use_labels = <Age_label>)
modalityB_h5ad.layers['input'] = modalityB_h5ad.X
aurora.configure_dataset(modalityB_h5ad, use_layers = 'input', use_uid = <sample_ID>,use_labels = <Age_label>)

modalityN_h5ad.layers['input'] = modalityN_h5ad.X
aurora.configure_dataset(modalityN_h5ad, use_layers = 'input', use_uid = <sample_ID>,use_labels = <Age_label>)

all_features = modalityA_h5ad.var_names.to_list() + modalityB_h5ad.var_names.to_list() + modalityN_h5ad.var_names.to_list()

Input:

• modalityX_h5ad: an AnnData object of modality X.
• <sample_ID>: the name of the unique sample_ID column in adata.obs.
• <Age_label>: the name of the numeric Age column in adata.obs.

Step 2: Train the model

aurora_model = aurora.fit_model(
    adatas = {"modality_A": modalityA_h5ad, 
              "modality_B": modalityB_h5ad,
              ...
              "modality_N": modalityN_h5ad}, 
    features = <features>,
    project_name = <project_name>
)
aurora_model.save("my_model.dill")

Input:

• adatas: a list of configured AnnData object.
• features: a list of all variable names of adatas, like: all_features
• project_name: the model will be saved in a folder named <project_name>. Default: my_aurora

Step 3: Generation with the model

import aurora
from aurora import load_model

#load model
aurora_model = load_model("my_aurora.dill")
#get sample embeddings from modality A
modalityA_h5ad.obsm["X_latent"] = aurora_model.encode_data("modality_A", modalityA_h5ad)
#generate all modality data
modality =aurora_model.net.x2z.keys()
for modality_ in modality:
    modalityA_h5ad.obsm["X_"+modality_] = aurora_model.decode_data(modality_, modalityA_h5ad,"modality_A")

Important Notice: Limited Edition & Commercial Services

This repository hosts a basic, stripped-down version of AURORA intended solely for academic research, demonstration, and non-commercial testing.

Access to the complete official version, advanced functionalities, official Agent demonstrations, priority technical support, tailored training workshops, and custom software development is available exclusively through commercial licensing. Enterprise users can choose from flexible payment models, including subscription-based plans and pay-per-use services.

To discuss licensing options or request a full version, please contact: contact@beyondagetec.com

Cite AURORA:

Chen, J., et al. A Generative AI Framework Unifies Human Multi-omics to Model Aging, Metabolic Health and Intervention Response. Cell Metab. (2026)