ASAP_CC#
The ASAP_CC model is a peak regression model fine-tuned to cell type-specific regions for 21 cell types in the human cingulate cortex (CC).
After pretraining on all 1.36M consensus peaks, the model was fine-tuned to cell type-specific peaks, defined as regions with a Gini index 1 standard deviation above the mean across all regions (275K peaks). Peak heights were calculated by taking mean coverage in the central 1000bp of resized peaks, normalized across cell types using min-max normalization with the top 3% of constitutive peaks per cell type. 2114bp one-hot encoded DNA sequences (centered on consensus peak summits) were used to predict mean peak accessibilities across cell types. Train/validation/test split was done by chromosome (chr2 and chr22 = validation, chr4 and chr9 = test).
The model is a CNN multiclass regression model using the dilated_cnn() architecture.
Details of the data and the model can be found in the original publication.
Citation
Sigalova, O.M., Pančíková, A., De Man, J., Theunis, K. et al. Modeling cis-regulatory variation in human brain enhancers across a large Parkinson’s Disease cohort. bioRxiv (2026). https://doi.org/10.64898/2026.03.15.711881
Usage#
1import crested
2import keras
3
4# download model
5model_path, output_names = crested.get_model("ASAP_CC")
6
7# load model
8model = crested.utils.load_model(model_path)
9
10# make predictions
11sequence = "A" * 2114
12predictions = crested.tl.predict(sequence, model)
13print(predictions.shape)