========
Outputs
========

The tool creates several files and folders in the specified output directory.
Below is the list and description of each output generated by the tool.

Training
---------

- ``model_final.pt``:
    The trained model.

- ``std.txt``:
    Standard deviation values for a given training data based on the specified z-score method ('zscore' and 'robustz').

    .. code-block::

        GDSC1	0.0	1.0
        GDSC2	0.0	1.0

- ``training_process.tsv``:
    Progress of training in each epoch.

    .. code-block::

        epoch	train_corr	train_loss	true_auc	pred_auc	val_corr	val_loss	grad_norm	elapsed_time
        0	0.0300	40.2999	0.7485	0.0637	nan	5.0000	0.0036	3.5948
        1	-0.0034	40.3000	0.7476	0.0652	0.0921	5.0000	0.0253	3.3162
        2	0.0722	40.2994	0.7477	0.0750	0.1169	5.0000	0.2334	3.2454
        3	0.1603	40.2977	0.7497	0.0976	0.1607	5.0000	0.4573	3.3437


Prediction
-----------

- ``predict.txt``:
    File containing prediction results.

    .. code-block::

        8.3823e-01
        7.6122e-01
        7.3318e-01

- ``predict_feature_grad_0.txt`` (``predict_feature_grad_1.txt``, ``predict_feature_grad_2.txt``):
    Files containing the gradients for each feature.

    .. code-block::

        -3.5083e-04	-1.5043e-03	1.4109e-02	4.2064e-04	-2.6275e-02 ..
        7.7715e-03	-2.1520e-02	-5.5406e-03	-1.3177e-03	-4.9724e-05 ..
        9.3111e-03	-2.0868e-02	3.6902e-05	6.6282e-03	-3.1622e-03 ..

- ``hidden`` directory:
    Directory with files containing the gradients of the hidden layer outputs.

- ``rlipp.out``:
    Output file with interpretation of predictions made by VNN.

    .. warning::

        The P_rho score is correlation coefficient. Generally 1 (positive values) mean positive correlation,
        0 no correlation, and -1 (negative values) negative correlation. However, uf training was not sufficient,
        the system's important scores like P_rho can be negative, which not necessarily mean negative correlation, but
        just the fact that the model has not learned meaningful patters from the data, and the hidden values are mostly
        random, which can result in arbitrary correlation. Recommended: increase number of epochs for training

    .. code-block::

        Term	P_rho	P_pval	C_rho	C_pval	RLIPP
        0	1.000e+00	0.000e+00	9.951e-01	0.000e+00	1.005e+00
        1	7.716e-01	5.312e-67	3.923e-01	1.064e-13	1.967e+00
        2	5.519e-01	6.151e-28	4.664e-01	2.182e-19	1.183e+00
        3	7.867e-01	2.638e-71	7.438e-01	7.026e-60	1.058e+00

- ``gene_rho.out``:
    Output file with Spearman correlation between gene embeddings and predicted AUC.

    .. code-block::

        Gene	Rho	P_val
        ABCB1	-1.215e-01	2.667e-02
        ABCC3	-9.125e-03	8.682e-01
        ABL1	5.741e-02	2.962e-01
        ABL2	-5.068e-02	3.565e-01

Annotation
-----------

- ``hierarchy.cx2``:
    File with hierarchy in HCX format annotated with interpretation results that will help determine importance of
    the subsystems in the hierarchical network.

- ``rlipp.out``:
    Aggregated interpretation scores of each provided RO-crates with prediction and interpretation results.

    .. code-block::

        Term	P_rho	P_pval	C_rho	C_pval	RLIPP	Disease
        0	1.00000e+00	0.00000e+00	9.95100e-01	0.00000e+00	1.00500e+00	unspecified
        1	7.71600e-01	5.31200e-67	3.92300e-01	1.06400e-13	1.96700e+00	unspecified
        2	5.51900e-01	6.15100e-28	4.66400e-01	2.18200e-19	1.18300e+00	unspecified

Logs and Metadata
-----------------

- ``output.log``:
    A standard log file recording events, errors, and other messages during the execution of the tool.

- ``error.log``:
    A specialized log file recording only error messages encountered during the execution of the tool.