STEAM_evolution_naive#
The STEAM_evolution_naive model is the mouse-only baseline from the evolutionary transfer learning study. It predicts cell-class-specific chromatin accessibility across 36 cell classes of the developing mouse embryo (E10–P0), trained solely on mouse chromatin accessibility data without any evolutionary information.
It is provided as the starting point of the STEAM progression (mouse-only → evolution-aware filtering → synteny-augmented STEAM_v1). 2114bp one-hot encoded DNA sequences are used to predict normalized Tn5 cut-site accessibility (log-scaled) over the central region of each window, per cell class.
The model is a CNN multiclass regression model using the dilated_cnn() architecture. Like the other STEAM models, it maps DNA sequence directly to accessibility and is genome-agnostic at inference (applicable to any mammalian genome).
Details of the data and the model can be found in the original publication.
Citation
Qiu, C., Daza, R.M., Welsh, I.C. et al. Evolutionary transfer learning enables organism-wide inference of mammalian enhancer landscapes (2026). https://doi.org/10.62329/hxkk6249
Usage#
1import crested
2import keras
3
4# download model
5model_path, output_names = crested.get_model("STEAM_evolution_naive")
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)