Contributing containers

We encourage you to explore the GAME modules page and begin with an Evaluator or Predictor codebase that most closely matches your own implementation.

The scripts listed below are meant to be re-usable across all GAME modules with minimal changes. We suggest testing your Evaluators/Predictors with the test module containers to validate their functionality.

Checklist for Predictor Modules

File	Description	Edits Required
config.py	Sets the Predictor base name, automatically appends the container build timestamp for module-level versioning, determines container vs. dev paths, and configures supported request/response wire formats.	Edits required for each Predictor.
predictor_RestAPI.py	Includes GET /formats and /help, POST /predict methods; validates/preprocesses requests.	Edits required.
predictor_content_handler.py	Decodes requests, encodes responses (JSON/MsgPack), adds predictor name, handles MIME negotiation, integrates with Flask.	No edits required if dealing with JSON and MsgPack.
schema_validation.py	Validates payload keys/values; preprocesses sequences for inference; provides progress feedback.	Edits required for predictor-specific sections.
error_checking_functions.py	Error classes (APIError, BadRequestError, etc.); validation functions for sequences, keys, readouts, ranges, and consistency.	No edits required.
<model_wrapper_script>	Model specific code that sets up a callable for predictor_RestAPI.py. Like predict_dream_rnn(), predict_borzoi(), etc.	Edit to define model specific code for each Predictor.
predictor_help_message.json	HELP file per GAME API spec. Must include game_schema_version (see Predictor Help Classes)	Edits required for each Predictor.
*_predictor.def	Apptainer definition file for building the container. The build-date label is used automatically by config.py for module-level versioning.	Edits required for each Predictor.

Predictor Responsibilities

• Set up endpoints for /formats, /help and /predict

• Validate the requests, decline any requests the model can’t fulfill

• Pre-process the data to add any adapters, and crop sequences if prediction ranges are sent

• Send sequences to the model, use Matcher if necessary

• Format predictions in API return formats, return HTTP error codes if necessary

Checklist for Evaluator Modules

File	Description	Edits Required
config.py	Sets the Evaluator base name, automatically appends the container build timestamp for module-level versioning, determines container vs. dev paths, and configures API communication settings (request/response format, retries, retry interval).	Edits required for each Evaluator.
evaluator_RestAPI.py	Loads/ validates input, sends data to predictor via HTTP, handles responses, saves raw predictions, computes metrics on success.	Minor edits based on Evaluator’s Data
evaluator_content_handler.py	Sends HTTP requests with retries, negotiates formats (JSON/MsgPack), deserializes responses safely.	No edits required
data_loader.py	Loads and validates input files (.json, .msgpack, .mpk), checks for duplicates, returns validated dictionary.	Edit to load datasets
evaluator_metrics_calculator.py	Calculates/saves correlation and cell-type specificity metrics, handles invalid/missing data, outputs CSV with timestamps.	Edits required for evaluation metric calculation
*evaluator.def	Apptainer definition file for building the container. Must include game_schema_version in the %labels block.	Edits required for each Evaluator.

Evaluator Responsibilities

• Parsing data from the Evaluator data folder and transforming into one of the Predictor defined request formats

• Issue POST /predict request to the Predictors

• Receive and save predictions

• Calculate evaluation metrics and save in required format