SeqDes model

get_pdb_data(pose::Pose, selection::Opt{AbstractSelection} = nothing)

Returns all PDB data necessary for calc_seqdes from the given Pose pose. If an AbstractSelection selection is provided, consider only the selected Atom instances (for Residue level data, since the selection is promoted, any Residue with at least 1 selected Atom is considered).

Should return:

• atom_coords: N x 3 Vector (atomic positions)
• atom_data: N x 4 Vector(atomic data):
  • [residue_id, bb_ind, atom_type, residue_type], where residue_id is 0-indexed; bb_ind is either 1 or 0, if the corresponding atom is in the protein backbone or not, respectively; atom_type is the index of the atom element in the Calculators.SeqDes.atom_types dictionary; and residue_type is the index of the residue type in the Calculators.SeqDes.residue_types dictionary.
• residue_bb_index: Nr x 4 (atomic index of all backbone atoms + CB)
  • [N, CA, C, CB], where all atomic indexes are 0-indexed; for residues without CB, the -1 index is used.
• residue_data: Dict('chain_code' => Nr x 4) (residue data)
  • [residue_id, residue_id_code, residue_index, residue_type], where residue_id_code can be ignored and set to ' ', residue_index is 0-indexed and residue_type is the index of the residue type in the Calculators.SeqDes.residue_types dictionary.
• residue_label: Nr x 1 (list of residue types)
• chis: Nr x 2 x 4 (chi dihedral angle values)
  • [[chi1_v, chi2_v, chi3_v, chi4_v], [chi1, chi2, chi3, chi4]], where chi1_v, chi2_v, etc is the actual dihedral value (in a 4 element array) and chi1, chi2, etc is either 1 or 0, if the corresponding chi dihedral exists or nor, respectively.

See also

calc_seqdes

Examples

julia> ProtoSyn.Peptides.Calculators.SeqDes.get_pdb_data(pose)
(...)

calc_seqdes([::Type{ProtoSyn.CUDA_2}], pose::Pose, selection::Opt{AbstractSelection}, [update_forces::Bool = false]; [use_cuda::Bool = true])

Calculates the Pose pose energy in accordance to the SeqDes machine learning model (see https://www.nature.com/articles/s41467-022-28313-9). In sum, this model evaluates the current sequence and chi dihedrals in accordance to a trained model on many natural proteins (but generalizable). If an AbstractSelection selection is given, only the selected Atom instances are considered for the energy calculation. This method does not calculate forces. As such, the update_forces flag has no effect and serves only for standardization purposes between Calculators methods. Currently, this method requires the usage of ProtoSyn.CUDA_2 acceleration type (and setting the use_cuda flag to true). Returns the energy value, nothing (forces), energy per residue and the calculated logits. Note that the calculated logits can be employed in the SeqDes ML model to retrieve rotamers according to the model. Both the energy per residue and logits are returned as PyObject tensors.

See also

get_default_seqdes

Examples

julia> ProtoSyn.Peptides.Calculators.SeqDes.calc_seqdes(pose, nothing)
(13.240120887756348, nothing, PyObject tensor(...), PyObject tensor(...))

get_default_seqdes(;[α::T = 1.0]) where {T <: AbstractFloat}

Return the default SeqDes energy EnergyFunctionComponent. α sets the component weight (on an EnergyFunction instance, 1.0 by default). This function employs calc_seqdes as the :calc function.

Settings

• use_cuda::Bool - Whether to use CUDA acceleration type. Currently, this EnergyFunctionComponent requires use_cuda to be true.

See also

calc_seqdes

Examples

```jldoctest julia> ProtoSyn.Peptides.Calculators.SeqDes.getdefaultseqdes() 🞧 Energy Function Component: +–––––––––––––––––––––––––-+ | Name | SeqDesMLModel | | Alpha (α) | 1.0 | | Update forces | false | | Calculator | calcseqdes | +–––––––––––––––––––––––––-+ | +–––––––––––––––––––––––––––––––––––––––––+ ├── ● Settings | Value | | +–––––––––––––––––––––––––––––––––––––––––+ | | usecuda | true | | +–––––––––––––––––––––––––––––––––––––––––+ | └── ○ Selection: nothing