DCGP model checkpoints
DCGP (Domestication Convergence Gene Prediction) predicts candidate genes under convergent selection during plant domestication. This repository contains 44 trained PyTorch classifier checkpoints. The classifiers use sequence embeddings from Evo 2. See the DCGP code repository for preprocessing, inference, and environment setup.
Download and run
Each checkpoint is stored as model/<model-id>/<model-id>.pt. Select the
model ID for your analysis. For example, from the root of the DCGP code repository:
hf download AIBreeding/DCGP model/Brassica_napusA/Brassica_napusA.pt --local-dir .
python embedding.py --fasta sequence/sample.fa --step 250 --window 1024 \
--layer blocks.24 --out sequence/sa.pt
python predict.py --data sequence/sa.pt \
--model model/Brassica_napusA/Brassica_napusA.pt --out results/pred
The embedding step uses evo2_7b and requires CUDA and SeqKit. The predictor
accepts precomputed embeddings and writes scores and predicted labels to
results/pred/predict_result.csv. For reproducible runs, pass the full Hub
commit SHA using hf download --revision. File hashes are listed in
SHA256SUMS.
Intended use
These checkpoints are intended for noncommercial research and reproduction of the DCGP inference workflow. Predicted scores can help prioritize candidates for follow-up analysis and experimental validation.
License
The DCGP checkpoints are available under Creative Commons Attribution-NonCommercial 4.0 International. The source code and upstream Evo 2 weights have their own license terms.