A Linear layer with built-in nonstandard initializations. Called just like torch.nn.Linear. Implements the initializers in 1.11.4, plus

A 3D rotation. Depending on how the object is initialized, the rotation is represented by either a rotation matrix or a quate

A Linear layer with built-in nonstandard initializations. Called just like torch.nn.Linear. Implements the initializers in 1.11.4, plus

A class representing a rigid transformation. Little more than a wrapper around two objects: a Rotation object and a [*, 3] translatio

Convenience class for rowwise dropout as described in subsection 1.11.6.

Standard multi-head attention using AlphaFold's default layer initialization. Allows multiple bias vectors.

Protein structure representation.

Standard multi-head attention using AlphaFold's default layer initialization. Allows multiple bias vectors.

Convenience class for columnwise dropout as described in subsection 1.11.6.

Implements Algorithm 15.

Implements Algorithm 12.

Implements Algorithm 11.

Returned by the parse function. Contains: mmcif_object: A MmcifObject, may be None if no chain could be successfully parsed.

Returned by the parse function. Contains: mmcif_object: A MmcifObject, may be None if no chain could be successfully parsed. errors:

Protein structure representation.

An error indicating that the query can't be aligned to the template.

Combines dense layer and scaling for swish activation. Parameters ---------- units: int Output embedding size.

Implements Algorithm 7.

Feed-forward network applied to MSA activations after attention. Implements Algorithm 9

Embeds the "template_angle_feat" feature. Implements Algorithm 2, line 7.

A naive fixup for misssing msa block Cross information of z,template_pair by point wise attention Cross information of s,template_angle b

Embeds "template_pair_feat" features. Implements Algorithm 2, line 9.

Implements Algorithm 17.

Implements Algorithm 14.

Implements Algorithm 13.

An error indicating that the hit align ratio to the query was too small.

Implements Algorithm 20, lines 11-14

Implements part of Algorithm 23.

Implements part of Algorithm 23.

Pytorch pickle loading workaround. https://github.com/pytorch/pytorch/issues/16797

An error indicating that a CA atom distance exceeds a threshold.

An error indicating that the hit date was after the max allowed date.

Computes a distogram probability distribution. For use in computation of distogram loss, subsection 1.9.8

An error indicating that the hit was an exact subsequence of the query.

Main Evoformer trunk. Implements Algorithm 6.

For use in computation of "experimentally resolved" loss, subsection 1.9.10

Almost identical to the standard EvoformerBlock, except in that the ExtraMSABlock uses GlobalAttention for MSA column attention and

Embeds unclustered MSA sequences. Implements Algorithm 2, line 15

Implements Algorithm 18.

Embeds a subset of the input features. Implements Algorithms 3 (InputEmbedder) and 4 (relpos).

Implements Algorithm 22.

Implements Algorithm 22.

An error indicating that the hit was too short.

Implements Algorithm 8. By rights, this should also be a subclass of MSAAttention. Alas, most inheritance isn't supported by TorchScript

Implements an Axial MSA Transformer block.

Implements an Axial MSA Transformer block.

Feed-forward network applied to MSA activations after attention. Implements Algorithm 9

For use in computation of masked MSA loss, subsection 1.9.9

Representation of a parsed mmCIF file. Contains: file_id: A meaningful name, e.g. a pdb_id. Should be unique amongst all files bein

Representation of a parsed mmCIF file. Contains: file_id: A meaningful name, e.g. a pdb_id. Should be unique amongst all files being proc

An error indicating that multiple chains were found for a given ID.

An error indicating that template mmCIF didn't contain atom positions.

An error indicating that template mmCIF didn't have any chains.

Optim wrapper that implements rate.

Optim wrapper that implements rate.

Implements the stochastic filters applied during AlphaFold's training. Because samples are selected from constituent datasets randoml