ValidationMetrics

Contains classes for evaluation of models performance on both docking tasks.

RMSD for interaction pose (IP) prediction. Accuracy, precision, recall, F1-score, and MCC (Matthews Correlation Coefficient) for fact-of-interaction (FI).

class Dock2D.Utility.ValidationMetrics.RMSD(ligand, gt_rot, gt_txy, pred_rot, pred_txy)

__init__(ligand, gt_rot, gt_txy, pred_rot, pred_txy)

calc_rmsd()

Returns

RMSD of predicted versus ground truth shapes.

get_XC()

Analog of inertia tensor and center of mass for rmsd calc. Where X is the inertia tensor, W is the sum of the bulk shape, and C is the center of mass.

Returns

\[\frac{2*X}{W}, \frac{C}{W}\]

class Dock2D.Utility.ValidationMetrics.APR

__init__()

Initialize epsilon to avoid division by ~zero.

calc_APR(data_stream, run_model, epoch=0, deltaF_logfile=None, experiment=None, stream_name=None)

Calculate accuracy, precision, recall, F1-score, and MCC (Matthews Correlation Coefficient) based on confusion matrix values during fact-of-interaction model evaluation.

Parameters

• data_stream – data stream

• run_model – run_model() passed from FI model

• epoch – current evaluation epoch

• deltaF_logfile – free energies per example and F_0 logfile

• experiment – current experiment name

• stream_name – name of current data stream

Returns

accuracy, precision, recall, f1score, mcc