Safety wrapper for PRNG keys.

Multihead attention.

Affine transformation represented by quaternion and vector.

Rigid Transformation, i.e. element of special euclidean group.

Triangle Attention. Jumper et al. (2021) Suppl. Alg. 13 "TriangleAttentionStartingNode" Jumper et al. (2021) Suppl. Alg. 14 "TriangleAttentionEnd

Triangle multiplication layer ("outgoing" or "incoming"). Jumper et al. (2021) Suppl. Alg. 11 "TriangleMultiplicationOutgoing" Jumper et al. (2021

random key generator

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

Given input reprensentation and frame produces points in global frame.

Transition layer. Jumper et al. (2021) Suppl. Alg. 9 "MSATransition" Jumper et al. (2021) Suppl. Alg. 15 "PairTransition"

Minimal AbLang gradient wrapper (VHH via AbLang1, scFv via AbLang2).

Single iteration (block) of Evoformer stack. Jumper et al. (2021) Suppl. Alg. 6 "EvoformerStack" lines 2-10

Protein structure representation.

Rot3Array Matrix in 3 dimensional Space implemented as struct of arrays.

Safety wrapper for PRNG keys.

Individual trajectory data container with metrics and metadata management. This class encapsulates all information associated with a single d

A single recycling iteration of AlphaFold architecture. Jumper et al. (2021) Suppl. Alg. 2 "Inference" lines 3-22

A single recycling iteration of AlphaFold architecture. Computes ensembled (averaged) representations from the provided features. These represent

Implements an Axial MSA Transformer block.

Callable to convert an unprocessed (labels + strings) batch to a processed (labels + tensor) batch.

Compute self-attention over columns of a 2D input.

Performs symmetrization, apc, and computes a logistic regression on the output features

A custom class that wraps a frozen IgLM model but uses an external seq_logits tensor for the variable (designed) region.

Embeds the input data and runs Evoformer. Produces the MSA, single and pair representations. Jumper et al. (2021) Suppl. Alg. 2 "Inference" line

Embeds the input data and runs Evoformer. Produces the MSA, single and pair representations.

A single iteration of iterative folding. First, each residue attends to all residues using InvariantPointAttention. Then, we apply transition lay

A single iteration of the main structure module loop. Jumper et al. (2021) Suppl. Alg. 20 "StructureModule" lines 6-21 First, each residue atten

Global attention. Jumper et al. (2021) Suppl. Alg. 19 "MSAColumnGlobalAttention" lines 2-7

Main I/O handler for Germinal design runs with thread-safe operations. This class provides comprehensive I/O management for Germinal design r

Invariant point attention module. The high-level idea is that this attention module works over a set of points and associated orientations in 3D

Invariant Point attention module. The high-level idea is that this attention module works over a set of points and associated orientations in 3D

MSA per-column attention. Jumper et al. (2021) Suppl. Alg. 8 "MSAColumnAttention"

MSA per-column global attention. Jumper et al. (2021) Suppl. Alg. 19 "MSAColumnGlobalAttention"

MSA per-row attention biased by the pair representation. Jumper et al. (2021) Suppl. Alg. 7 "MSARowAttentionWithPairBias"

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

Class to make side chain atoms.

Class to make side chain atoms.

Computes mean outer product. Jumper et al. (2021) Suppl. Alg. 10 "OuterProductMean"

Module for projecting Rigids via a quaternion.

Compute self-attention over rows of a 2D input.

Embeds a single template. Jumper et al. (2021) Suppl. Alg. 2 "Inference" lines 9+11

Embed a single template.

Embeds a set of templates. Jumper et al. (2021) Suppl. Alg. 2 "Inference" lines 9-12 Jumper et al. (2021) Suppl. Alg. 17 "TemplatePointwiseAttenti

Embed a set of templates.

Single Iteration of Template Embedding.

Class representing a template hit.

Pair stack for the templates. Jumper et al. (2021) Suppl. Alg. 16 "TemplatePairStack"

Vec3Array in 3 dimensional Space implemented as struct of arrays. This is done in order to improve performance and precision. On TPU small matrix

_LayerStack impl with no per-layer state provided to the function.

_LayerStack impl with per-layer state provided to the function.

AlphaFold Jumper et al. (2021) Suppl. Alg. 2 "Inference

AlphaFold-Multimer model with recycling.

Runs the alignment tools and assembles the input features.

Head to predict a distogram. Jumper et al. (2021) Suppl. Sec. 1.9.8 "Distogram prediction"

Predicts if an atom is experimentally resolved in a high-res structure. Only trained on high-resolution X-ray crystals & cryo-EM. Jumper et al. (

LayerNorm module. Equivalent to hk.LayerNorm but with different parameter shapes: they are always vectors rather than possibly higher-rank tensors

Protein folding specific Linear module. This differs from the standard Haiku Linear in a few ways: * It supports inputs and outputs of arbitrary