CAFA Predictions Evaluation

Test data are available in FigShare: https://figshare.com/articles/dataset/cafa_test_data_tgz/16713046

This repository is a collection of code for evaluating CAFA predictions

See also the walkthrough example

To use the code, the steps are:

1. Obtain or generate propagation Pandas DataFrame files based on the relevant CAFA challenge obo file. These files are pickled DataFrames, one per relevant GO ontology namespace that are used for propagating both benchmark and prediction annotation data. To generate these files yourself, see generate_propagation_map_dataframe.py. This file can be called directly from the shell with a yaml configuration file. See obo_parser_config.yml as an example of the expected configuration.

2. Obtain or generate weighted DAG files (one per GO ontology namespace). These are pickled Pandas DataFrame files containing the Information Content data for the relevant DAG nodes. To generate your own weighted DAG files, see generate_information_content_matrices.py. This file uses the same configuration yaml as step 1.

3. Parse raw benchmark data into json-formatted per-species, per-ontology (CCO, MFO, BPO, etc) files. This step uses raw_benchmark_parser.py. This step requires a yaml configuration file as well. See parser_config.yml for an example of the necessary configuration keys and values

4. Parse raw prediction data into json-fomatted files on a per-species and per-ontology basis. This step uses raw_prediction_parser.py in conjunction with the same configuration file from step #3.

5. Evaluate the prediction data on a per-species, per-ontology basis against the benchmark data using evaluate_species_prediction.py

6. Generate cross-species evaluation metrics using evaluate_cross_species.py