Skip to content

module adabmDCA.api.splitting

High-level API for homology-aware alignment splitting.

function split_alignment

split_alignment(
    alignment: 'AlignmentInput',
    method: 'str' = 'clustering',
    identity: 'float' = 0.8,
    train_fraction: 'float' = 0.8,
    t1: 'float' = 0.5,
    t2: 'float' = 0.5,
    t3: 'float' = 1.0,
    max_train: 'int | None' = None,
    max_test: 'int | None' = None,
    attempts: 'int' = 1,
    alphabet: 'str' = 'protein',
    seed: 'int' = 0,
    device: 'str' = 'auto'
) → ProfileSplitResult

Split an alignment into training and test sets that are not too similar.

method="clustering" clusters sequences at identity (MMseqs2 when installed and there are at least 100 sequences, otherwise a built-in greedy clustering) and assigns whole clusters to the training set until about train_fraction of the sequences are in it. method="cobalt" runs the Cobalt algorithm (Petti and Eddy, 2022), whose thresholds t1-t3 bound the identities between and within the two sets.

Args:

  • alignment: Alignment to split (path, Alignment or sequences).
  • method: "clustering" or "cobalt".
  • identity: Clustering: sequence identity, in (0, 1], that joins two sequences into one cluster.
  • train_fraction: Clustering: target fraction of sequences in the training set.
  • t1: Cobalt: no test sequence is more than t1 identical to a training sequence.
  • t2: Cobalt: no two test sequences are more than t2 identical.
  • t3: Cobalt: no two training sequences are more than t3 identical.
  • max_train: Cobalt: largest training set, or None for no limit.
  • max_test: Cobalt: largest test set, or None for no limit.
  • attempts: Cobalt: random attempts; the one maximizing len(training) * len(test) is kept.
  • alphabet: "protein", "rna", "dna" or a custom token string.
  • seed: Random seed.
  • device: "auto", "cpu", "cuda" or "mps" for identity computations.

Returns: A ProfileSplitResult with the two alignments.

Raises:

  • InputValidationError: If a setting is out of range or the alignment has invalid sequences.
  • ConvergenceError: If Cobalt finds no split with both sets non-empty.

Example: 

split = split_alignment("family.fasta", alphabet="rna", identity=0.8)
split.save_bundle("splits/family")

This file was automatically generated via lazydocs.