Get started by using pre-built GAME modules

GAME modules can be run interactively by users or using our submission scripts [LINK]. To parallelize predictions you can use Predictor Distributor .

Predictor run command:

apptainer run --containall predictor.sif HOST PORT

To use a container with NVIDIA GPU:

apptainer run --containall --nv predictor.sif HOST PORT

To use a container with GPU + Matcher:

apptainer run --containall --nv predictor.sif HOST PORT MATCHER_HOST MATCHER_PORT

NOTE: For AMD GPUs and ROCm framework, please refer to Apptainer’s Documentation.

Evaluator run command:

apptainer run --containall \
    -B /path/to/evaluator_data:/evaluator_data  \
    -B /path/to/predictions:/predictions  \
    evaluator.sif HOST PORT /predictions

Updated list of current GAME modules can be found here: [LINK]

Running the DREAM-RNN container (Matcher not required) with a sample dataset

To run a test prediction using the DREAM-RNN container and sample Evaluator container:

1. Download the containers from Zenodo: [LINK]

   mkdir DREAMRNN
   mkdir test_evaluator
   

   cd DREAMRNN
   wget -O dream_rnn_predictor.sif [LINK]
   

   cd test_evaluator
   wget -O evaluator.sif [LINK]
   wget -O evaluator_data.zip [LINK]
   unzip evaluator_data.zip
   mkdir predictions
   

   Note: if you run into issues downloading the evaluator_data folder you may need to manually download it off Zenodo.

2. Get the IP Address of where the Predictor is running

   Note: PORTs above 1024 are usually free to use

   hostname -I (NOTE: This could be different for different HPC platforms – -I, -i, no flag, etc.)

3. Start the DREAMRNN Predictor with the IP address and PORT arguments

   apptainer run --containall --nv predictor.sif HOST PORT

   Example: apptainer run --containall --nv predictor.sif 172.16.47.243 5000

4. Start the test Evaluator

   apptainer run --containall \
       -B /path/to/evaluator_data:/evaluator_data  \
       -B /path/to/predictions:/predictions  \
       evaluator.sif HOST PORT /predictions
   

   Example:

   apptainer run --containall \
       -B /path/to/evaluator_data:/evaluator_data  \
       -B /path/to/predictions:/predictions  \
       evaluator.sif 172.16.47.243 5000 /predictions
   

   The -B mounts local directories so that the Evaluator container can read in the JSON file from a local folder and write the prediction to the locally created /predictions folder.

5. If the Evaluator-Prediction communication was successful a JSON file will be found in the predictions/ folder.

Yay! You just completed a successful communication between the DREAMRNN model and a test sequence set with GAME :)

{
    "predictor_name": "DREAM-RNN_Human_K562",
    "prediction_tasks": [
        {
            "name": "gosai_synthetic_sequences",
            "type_requested": "expression",
            "type_actual": ["expression"],
            "cell_type_requested": "K562",
            "cell_type_actual": "K562",
            "scale_prediction_requested": "log",
            "scale_prediction_actual": "log",
            "species_requested": "homo_sapiens",
            "species_actual": "homo_sapiens",
            "predictions": {
                "7:70038969:G:T:A:wC": [
                    -0.4900762140750885
                ],
                "1:192696196:C:T:A:wC": [
                    -0.4205487370491028
                ],
                "1:211209457:C:T:A:wC": [
                    -0.2514425814151764
                ],
                "15:89574440:GT:G:A:wC": [
                    1.1541708707809448
                ],
                "15:89574440:GT:G:R:wC": [
                    1.1637296676635742
                ]
            }
        }
    ]
}

Running Enformer container (with Matcher) with a sample dataset

Coming soon