DCAF

Data-Criterion-model Associated Framework for full cycle of single-cell pertubation modeling!

DCAF-data & DCAF-criterion

Installation of DCAF-data and DCAF-criterion

git clone https://github.com/BGIResearch/DCAF.git

cd DCAF

conda create -n DCAF_data_cri python=3.9

pip install .

pip install -r requirements.txt

Activate environment for DCAF-data and DCAF-criterion

conda activate DCAF_data_cri

Demo for DCAF-data

python run_recover.py -m demo

Dataset Tian, derived from paper is applied for this demonstration.

Use DCAF-data to recover control and case status of your own data

python run_recover.py -ctrl NPZ_FULL_PATH_OR_ITS_FILE_NAME_UNDER_/data
                      -case NPZ_FULL_PATH_OR_ITS_FILE_NAME_UNDER_/data

Control and case npz files shall be placed in the same path. They don't necessarily be of the same size.

Demo for DCAF-criterion

python run_measure.py -m demo

Synthetic dataset using our article (a link to be inserted) is applied for this demonstration.

Use DCAF-criterion to assess accuracy of your own output, using your own dataset

python run_measure.py -ctrl NPZ_FULL_PATH_OR_ITS_FILE_NAME_UNDER_/data
                      -case NPZ_FULL_PATH_OR_ITS_FILE_NAME_UNDER_/data
                      -pred NPZ_FULL_PATH_OR_ITS_FILE_NAME_UNDER_/data

Control, case, and predicted npz files shall be placed in the same path. They shall be of the same size.

DCAF-model

Hardware Requirement and Test Performance

Nvidia GPU with at least 8 GB VRAM is required, 16 GB is recommended.

Multi-card acceleration is not supported.

Test performance:

Item	Specification
CPU Architecture & Model	aarch64 (ARM), HiSilicon Kunpeng-920
GPU	NVIDIA A100-PCIE-40GB ($\times$ 1)
Batch size	4
No. of cells	50
Peak VRAM	11 GB
Time	9.2 s

Installation of DCAF-model

git clone https://github.com/BGIResearch/DCAF.git

cd DCAF

conda create -n DCAF_model python=3.9

conda activate DCAF_model

pip install -r requirements_model.txt

Activate environment for DCAF-model

conda activate DCAF_model

Demo for DCAF-model inference

Download fintuned models and dictionary, from drive to infer_by_ctrl/ours/data；

Use notebook for inference, at https://github.com/BGIResearch/DCAF/blob/main/infer_each_cell.ipynb