Usage

This page should provide information on how to use aixport

In a project

To use Artificial Intelligence Predictive Oncology Research Toolkit in a project:

import aixport

On the command line

For information invoke aixportcmd.py -h

Example usage

Training

$ aixportcmd.py train trainout --input input_train.txt --algorithms elasticnet_drecmd.py --run_mode bash

input_train.txt should contain absolute paths to *_train_rocrate directories, one per line. The command will populate trainout with bash script that populates trainout/trainedmodels with per-algorithm trained model directories. Then run the bash script:

./trainout/bash_train_job.sh

Prediction

$ aixportcmd.py predict predictout --input input_test.txt --trainedmodels trainout/trainedmodels --algorithms elasticnet_drecmd.py --run_mode bash

input_test.txt should list the *_test_rocrate directories with test data, and trainout/trainedmodels should contain subdirectories named <drugname>_train_rocrate_<algorithm> produced by the training step. Then run the bash script:

./predictout/bash_predict_job.sh

Benchmark

$ aixportcmd.py benchmark benchmarkout \
    --input_test_rocrates input_test.txt \
    --predictions_rocrate predictout

input_test.txt should match the file used during prediction and list the test RO-Crates, one per line. predictout should be the prediction RO-Crate folder produced by aixportcmd.py predict (or the pipeline), containing the predictions subdirectory. The benchmark command creates benchmarkout (must not already exist), computes Pearson/Spearman correlations between predictions and ground truth, writes results.csv, and generates results.png/results.svg for quick inspection.

Custom RO-Crate pipeline

$ bash scripts/run_custom_dataset.sh \
    --rocrates-dir /path/to/rocrates \
    --output-dir /path/to/custom_run \
    --model-config configs/custom_dataset_models.json

The --rocrates-dir input must already contain per-drug *_train_rocrate and *_test_rocrate folders.

By default, the script installs aixport and the enabled model packages with their declared Python dependencies before running. Use --skip-install only when the environment is already prepared.

The model config JSON supports:

enabled to turn a model on or off
optimize to control per-model hyperparameter optimization
install_path to install a custom model repo before running

Via Docker

Example usage

TODO: Add information about example usage

Coming soon ...

RO-Crate Zipping Utility

The ROCrateZipper helper can bundle an entire RO-Crate directory into a ZIP archive and inspect or extract its contents. This is useful for publishing or sharing generated RO-Crates.

Example usage

from aixport import ROCrateZipper

zipper = ROCrateZipper("/path/to/folder", "/path/to/output.zip")
zipper.zip_directory()
print(zipper.list_contents())  # View all files in the zip
data = zipper.read_file("subfolder/data.txt")  # Access a specific file
zipper.extract_file("subfolder/data.txt", "/tmp/extracted/")  # Extract one file