UtilityFunctions

class Dock2D.Utility.UtilityFunctions.UtilityFunctions(experiment=None, model_name=None)

check_model_gradients(model)

Check current model parameters and gradients in-place. Specifically if weights are frozen or updating

make_boundary(grid_shape, gaussian_blur_bulk=False)

Create the boundary feature for data generation and unit testing.

Parameters

grid_shape – input shape grid image

Returns

features stack with original shape as “bulk” and created “boundary”

orthogonalize_feats(scoring_weights, feat_stack)

Orthogonalize learned shape features for single shape.

Parameters

• scoring_weights – learned scoring coefficients from scoring function

• feat_stack – feature stack for one shape [bulk, boundary]

Returns

orthogonalized feature stack

plot_assembly(receptor, ligand, gt_rot, gt_txy, pred_rot=None, pred_txy=None, tiling=False, interaction_fact=False)

Plot the predicting docking pose for the IP task. From left to right, plots the ground truth docking pose, the pose passed into the model, and the predicted docking pose.

Parameters

• receptor – receptor shape grid image

• ligand – ligand shape grid image

• gt_rot – ground truth rotation

• gt_txy – ground truth translation [x, y]

• pred_rot – predicted rotation

• pred_txy – predicted translation [x, y]

Returns

plotting object with specified poses

plot_features(rec_feat, lig_feat, scoring_weights, plot_count=0, stream_name='trainset', model_name=None)

Plot the learned shape pair features (bulk and boundary) from the docking model from Docking in model_docking.py.

Parameters

• rec_feat – receptor feature stack

• lig_feat – ligand feature stack

• receptor – receptor shape grid image

• ligand – ligand shape grid image

• scoring_weights – learned scoring coefficients used in scoring function

• plot_count – plotting index used in titles and filename

• stream_name – data stream name

plot_predicted_pose(receptor, ligand, gt_rot, gt_txy, pred_rot, pred_txy, plot_count, stream_name)

Plotting helper function for plot_assembly().

Parameters

• receptor – receptor shape grid image

• ligand – ligand shape grid image

• gt_rot – ground truth rotation

• gt_txy – ground truth translation [x, y]

• pred_rot – predicted rotation

• pred_txy – predicted translation [x, y]

• plot_count – plotting index used in titles and filename

• stream_name – data stream name

plot_rotation_energysurface(fft_score, pred_txy, num_angles=360, stream_name=None, plot_count=0, plot_pub=False)

Plot the lowest energy translation index from fft_score per rotation angle as an energy surface curve.

Parameters

• fft_score – FFT scores generated using a docker method.

• pred_txy – predicted translation [x, y]

• num_angles – number of angles used to generate fft_score

• stream_name – data stream name

• plot_count – plotting index used in titles and filename

read_pkl(filename)

Parameters

filename – filename.pkl to load

Returns

data

rotate(repr, angle)

Rotate a grid image using 2D rotation matrix.

Parameters

• repr – input grid image

• angle – angle in radians

Returns

rotated grid image

rotate_gridligand(ligand, rotation_angle)

Rotate grid image in degrees using scipy.ndimage.rotate(), an alternative rotation function for plot_assembly()

Parameters

• ligand – grid image of ligand

• rotation_angle – angle in degrees

Returns

rotated ligand

swap_quadrants(input_volume)

FFT returns features centered with the origin at the center of the image, not at the top left corner.

Parameters

input_volume – FFT output array

translate_gridligand(ligand, tx, ty)

Translate grid image using scipy.ndimage.shift() for plot_assembly()

Parameters

• ligand – grid image of ligand

• tx – x dimension translation

• ty – y dimension translation

Returns

translated ligand

weights_init(model)

Initialize weights for SE(2)-equivariant convolutional network. Generally unused for SE(2) network, as e2nn library has its own Kaiming He weight initialization.

write_pkl(data, filename)

Parameters

• data – to write to .pkl file

• filename – specify filename.pkl