Transforms#

This section describes the configurations for various torch_geometric.transforms.Transforms objects used throughout the frame.

Note

These are typically not used in the CLI - they are rather primitives that we use in defining Tasks.

See also

Tasks

None#

This config file is used to specify no transforms.

transforms/none.yaml#

Data Transforms#

These transforms are used to modify the input data in some way, such as handling edge cases.

Remove Missing \(C_{\alpha}\) Atoms#

transforms/remove_missing_ca.yaml#

remove_missing_ca:
  _target_: proteinworkshop.tasks.remove_missing_ca.RemoveMissingCa
  fill_value: 1e-5 # Value used to indicate missing atoms
  ca_idx: 1 # Index of CA atoms in the AtomTensor

Generic Task Transforms#

Binding Site Prediction#

Protein Protein Site Prediction#

transforms/generic.yaml#

ppi_site_prediction:
  _target_: proteinworkshop.tasks.ppi_site_prediction.BindingSiteTransform
  radius: 3.5 #Maximum distance between chains to be considered as interacting
  ca_only: False #Whether to use only the alpha carbon atoms for determining interactions

Denoising Transforms#

Sequence Denoising#

transforms/sequence_denoising.yaml#

sequence_denoising:
  _target_: proteinworkshop.tasks.sequence_denoising.SequenceNoiseTransform
  corruption_rate: 0.25 # Fraction of residues to corrupt
  corruption_strategy: "mutate" # Whether to 'mutate' or 'mask'

Structure Denoising#

transforms/structure_denoising.yaml#

structure_denoising:
  _target_: proteinworkshop.tasks.structural_denoising.StructuralNoiseTransform
  corruption_rate: 0.1 # How much noise to apply
  corruption_strategy: "gaussian" # Whether to use a 'gaussian' or 'uniform' distribution

Torsional Denoising#

transforms/torsion_denoising.yaml#

torsional_denoising:
  _target_: proteinworkshop.tasks.torsional_denoising.TorsionalNoiseTransform
  corruption_rate: 0.1 # How much noise to apply
  corruption_strategy: "gaussian" # Whether to use a 'gaussian' or 'uniform' distribution