Runs alignment tools on a sequence

Run ESMFold on sequence.

Sets up inference run. All outputs are written to {output_dir}/{date_time} where {output_dir} is created at initializati

Generate samples from the target SO(3) distribution by sampling a rotation axis angle, which are then combined into a rotation vector

Uses the inverse cdf to sample an angle of rotation from IGSO(3). Args: t: time of Brownian motion n_samples: number

Create a series of samples from the IGSO(3) angle distribution. Args: sigma_indices (torch.Tensor): Indices of the IGSO3

Uses the inverse cdf to sample an angle of rotation from IGSO(3). Args: t: continuous time in [0, 1]. n_samples: numb

Sample MSA randomly, remaining sequences are stored are stored as `extra_*`.

Uniformly sample rotation axis for constructing the overall rotation. Args: num_sigma (int): Number of samples to draw f

Save a dictionary of tensors to the cache file. If overwrite is set to True, an existing file is overwritten, otherwise a warning is

Computes the score of IGSO(3) density as a rotation vector. Args: vec: [..., 3] array of axis-angle rotation vectors.

Convert all submodules whose types match one of those in the input list to recursively scripted equivalents in place. To script the entire

Run self-consistency on design proteins against reference protein. Args: decoy_pdb_dir: directory where designed protein

Calculates the index for discretized sigma during IGSO(3) initialization.

Detaches the underlying rotation object Returns: A transformation object with detached rotations

Implements Algorithm 27 (torsionAngleLoss) Args: angles_sin_cos: [*, N, 7, 2] predicted angles

Converts a `Protein` instance to a PDB string. Args: prot: The protein to convert to PDB. Returns: PDB string.

Graph-conditioned sequence model

Computes the sum along segments of a tensor. Similar to tf.unsorted_segment_sum, but only supports 1-D indices. :param data: A tensor w

Returns True if the atom will be restrained by the given restraint set.

Loss to penalize steric clashes within residues. This is a loss penalizing any steric violations or clashes of non-bonded atoms in a given pe

Function to be mapped over the ensemble dimension.

Serialize data into a pickle file.

Allows torch-style indexing over the virtual shape of the rotation object. See documentation for the shape property.

Indexes the affine transformation with PyTorch-style indices. The index is applied to the shared dimensions of both the rota

Args: rot_mats: A [*, 3, 3] rotation matrix tensor. Mutually exclusive with quats

Args: rots: A [*, 3, 3] rotation tensor trans: A corresponding [*, 3] translation tensor

Args: model: A torch.nn.Module whose parameters are to be tracked decay: A value (usu

Args: c_in: Input channel dimension c_hidden: Overall hidden channel dimension (not p

Args: c_q: Input dimension of query data c_k: Input dimension of key data

Args: c_z: Input channel dimension c: Hidden channel dimension

Args: c_m: MSA channel dimension c_z: Pair channel dimension c_hidden_msa

Args: c_hidden: Hidden channel dimension

Args: c_s: Single representation channel dimension c_z: Pair representation channel d

Args: c_s: Single representation channel dimension

Args: c_s: Single representation channel dimension c_z: Pair representation channel d

Args: c_m: MSA channel dimension c_z: Pair embedding channel dimension mi

Args: c_in: Final dimension of "template_angle_feat" c_out: Output channel dimension

Args: c_in: c_out: Output channel dimension

Args: c_in: Input channel dimension c_out: Output channel dimension

Args: r: Dropout rate batch_dim: Dimension(s) along which the dropout mask is shared

Args: c_m: Input channel dimension c_z: Pair embedding channel dimension

Args: c_m: MSA channel dimension c_hidden: Per-head hidden channel dimension

Args: c_z: Pair transition channel dimension n: Factor by which c_z is multiplied to