pallatom.helpers.atom_utils

Attributes

ATOM37_C
ATOM37_CA
ATOM37_CB
ATOM37_N
ATOM37_O
ATOM5_C
ATOM5_CA
ATOM5_CB
ATOM5_ELEMENTS
ATOM5_N
ATOM5_NAMES
ATOM5_O
PDB_CHAIN_IDS
PDB_MAX_CHAINS
atom_types
restype_1to3
restype_3to1
restype_num
restype_order
restypes
rigid_group_atom_positions

Classes

Protein

Protein structure representation.

Functions

atom37_to_atom5(→ tuple[jaxtyping.Float[torch.Tensor, ...)	Extract the 5 backbone+Cβ atoms from atom37 representation.
atom37_to_cb(→ tuple[jaxtyping.Float[torch.Tensor, ...)	Full pipeline: atom37 → atom5 → Cβ / pseudo-Cβ.
center_positions(np_example)	Center 'atom_positions' on CA center of mass.
get_cb_coords(→ tuple[jaxtyping.Float[torch.Tensor, ...)	Extract Cβ (slot 4) from atom5, replacing missing Cβ (Gly) with pseudo-Cβ.
make_fixed_size(np_example[, max_seq_length])	Pad features to fixed sequence length, i.e. currently axis=0.
make_np_example(coords_dict)	Make a dictionary of non-batched numpy protein features.
protein_from_pdb(→ Protein)	Parse a PDB file (ATOM records only) into a Protein using the atom37 layout.
pseudo_cb(→ jaxtyping.Float[torch.Tensor, ... 3])	Compute a virtual Cβ from backbone geometry (Gly-safe).
to_pdb(→ str)	Converts a Protein instance to a PDB string.

Module Contents

class pallatom.helpers.atom_utils.Protein

Protein structure representation.

aatype: jaxtyping.Int[numpy.ndarray, num_res]

atom_mask: jaxtyping.Float[numpy.ndarray, num_res num_atom_type]

atom_positions: jaxtyping.Float[numpy.ndarray, num_res num_atom_type 3]

b_factors: jaxtyping.Float[numpy.ndarray, num_res num_atom_type]

chain_index: jaxtyping.Int[numpy.ndarray, num_res]

residue_index: jaxtyping.Int[numpy.ndarray, num_res]

pallatom.helpers.atom_utils.atom37_to_atom5(atom37_positions: jaxtyping.Float[torch.Tensor, B N_res 37 3], atom37_mask: jaxtyping.Float[torch.Tensor, B N_res 37]) → tuple[jaxtyping.Float[torch.Tensor, B N_res 5 3], jaxtyping.Float[torch.Tensor, B N_res 5]]

Extract the 5 backbone+Cβ atoms from atom37 representation.

Returns:

• atom5_positions ((B, N_res, 5, 3))

• atom5_mask ((B, N_res, 5))

pallatom.helpers.atom_utils.atom37_to_cb(atom37_positions: jaxtyping.Float[torch.Tensor, B N_res 37 3], atom37_mask: jaxtyping.Float[torch.Tensor, B N_res 37]) → tuple[jaxtyping.Float[torch.Tensor, B N_res 3], jaxtyping.Bool[torch.Tensor, B N_res]]

Full pipeline: atom37 → atom5 → Cβ / pseudo-Cβ.

Returns:

• cb ((B, N_res, 3) — real Cβ where available, pseudo-Cβ otherwise)

• pseudo_beta_mask ((B, N_res) — True where real Cβ present, False where pseudo-Cβ was used)

pallatom.helpers.atom_utils.center_positions(np_example)

Center ‘atom_positions’ on CA center of mass.

pallatom.helpers.atom_utils.get_cb_coords(atom5_positions: jaxtyping.Float[torch.Tensor, B N_res 5 3], atom5_mask: jaxtyping.Float[torch.Tensor, B N_res 5], fill_pseudo: bool = True) → tuple[jaxtyping.Float[torch.Tensor, B N_res 3], jaxtyping.Bool[torch.Tensor, B N_res]]

Extract Cβ (slot 4) from atom5, replacing missing Cβ (Gly) with pseudo-Cβ.

Parameters:

• atom5_positions ((B, N_res, 5, 3))

• atom5_mask ((B, N_res, 5) — 1 where atom is present)

• fill_pseudo (if True, compute pseudo-Cβ wherever Cβ is absent)

Returns:

• cb ((B, N_res, 3) — real Cβ where available, pseudo-Cβ otherwise)

• pseudo_beta_mask ((B, N_res) — True where real Cβ present, False where pseudo-Cβ was used)

pallatom.helpers.atom_utils.make_fixed_size(np_example, max_seq_length=500)

Pad features to fixed sequence length, i.e. currently axis=0.

pallatom.helpers.atom_utils.make_np_example(coords_dict)

Make a dictionary of non-batched numpy protein features.

pallatom.helpers.atom_utils.protein_from_pdb(pdb_path: str) → Protein

Parse a PDB file (ATOM records only) into a Protein using the atom37 layout.

pallatom.helpers.atom_utils.pseudo_cb(n: jaxtyping.Float[torch.Tensor, ... 3], ca: jaxtyping.Float[torch.Tensor, ... 3], c: jaxtyping.Float[torch.Tensor, ... 3]) → jaxtyping.Float[torch.Tensor, ... 3]

Compute a virtual Cβ from backbone geometry (Gly-safe).

Uses the standard ideal-geometry recipe:

b = Cα - N (N→Cα bond vector) d = C - Cα (Cα→C bond vector) Cross them, then combine with ideal tetrahedral offsets.

This matches the AlphaFold2 / ESMFold convention exactly.

pallatom.helpers.atom_utils.to_pdb(prot: Protein) → str

Converts a Protein instance to a PDB string.

Parameters:

prot – The protein to convert to PDB.

Returns:

PDB string.

pallatom.helpers.atom_utils.ATOM37_C = 2

pallatom.helpers.atom_utils.ATOM37_CA = 1

pallatom.helpers.atom_utils.ATOM37_CB = 4

pallatom.helpers.atom_utils.ATOM37_N = 0

pallatom.helpers.atom_utils.ATOM37_O = 3

pallatom.helpers.atom_utils.ATOM5_C = 2

pallatom.helpers.atom_utils.ATOM5_CA = 1

pallatom.helpers.atom_utils.ATOM5_CB = 4

pallatom.helpers.atom_utils.ATOM5_ELEMENTS

pallatom.helpers.atom_utils.ATOM5_N = 0

pallatom.helpers.atom_utils.ATOM5_NAMES = ['N', 'CA', 'C', 'O', 'CB']

pallatom.helpers.atom_utils.ATOM5_O = 3

pallatom.helpers.atom_utils.PDB_CHAIN_IDS = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789'

pallatom.helpers.atom_utils.PDB_MAX_CHAINS = 62

pallatom.helpers.atom_utils.atom_types = ['N', 'CA', 'C', 'CB', 'O', 'CG', 'CG1', 'CG2', 'OG', 'OG1', 'SG', 'CD', 'CD1', 'CD2', 'ND1',...

pallatom.helpers.atom_utils.restype_1to3

pallatom.helpers.atom_utils.restype_3to1

pallatom.helpers.atom_utils.restype_num = 21

pallatom.helpers.atom_utils.restype_order

pallatom.helpers.atom_utils.restypes = ['A', 'R', 'N', 'D', 'C', 'Q', 'E', 'G', 'H', 'I', 'L', 'K', 'M', 'F', 'P', 'S', 'T', 'W', 'Y', 'V', 'X']

pallatom.helpers.atom_utils.rigid_group_atom_positions