This record provides a Predictor container for the DREAM-RNN model (Rafi et al. 2024). DREAM-RNN is a deep learning model optimized through the Prix Fixe framework, which was developed to systematically evaluate model architectures and training strategies from the Random Promoter DREAM Challenge. The DREAM-RNN Predictor can only return point expression predictions for K562 (Matcher, for automated task alignment, is never used, and it will return K562 predictions regardless of the cell type requested). Because it has hardcoded adapters (that were present in the model’s training data), it will ignore any adapter sequences that are sent. Sequences shorter than its 200bp input are centered and padded equally with Ns on either side, while sequence longer than 200bp are cropped to the target input length. Prediction ranges sent from the Evaluator are used to crop the input sequence to the desired start and stop indices. Running the container: Ensure Apptainer is intalled in the system the container is intended to run. Always run the Predictor first, so it can be reached at the bound host IP and port. The evaluator can then make requests to the exposed endpoints: apptainer run --nv --containall dream_rnn_predictor.sif HOST_IP HOST_PORT API Endpoints - GET /formats - Returns supported request and response formats - GET /help - Returns predictor metadata and usage information - POST /predict - Main endpoint for submitting prediction requests Additional information about GAME can be found on GitHub: Genomic API for Model Evaluation DREAM-RNN Predictor-specific information can be found here: DREAM-RNN K562 Code repository for the DREAM Challenge models: DREAM Challenge 2022 Github The paper can be found here: Rafi, A.M., Nogina, D., Penzar, D. et al. A community effort to optimize sequence-based deep learning models of gene regulation. Nat Biotechnol 43, 1373–1383 (2025). https://doi.org/10.1038/s41587-024-02414-w