SF-Cluster
Copyright (c) 2026 Hanqun Cao

This software builds on the following third-party work, which it depends on at
runtime but does NOT redistribute. Each retains its own license and authorship.

1. FrustrAI-Seq — per-residue frustration prediction model.
   License: Apache-2.0.
   Source:  https://github.com/leuschjanphilipp/FrustrAI-Seq
   Weights: https://huggingface.co/leuschj/FrustrAI-Seq
   SF-Cluster calls this model to compute the frustration features that guide
   MSA subsampling. Install and weights are fetched from the upstream sources.

2. ProtT5 (Rostlab/prot_t5_xl_half_uniref50-enc) — protein language model used
   internally by FrustrAI-Seq.
   Source:  https://huggingface.co/Rostlab/prot_t5_xl_half_uniref50-enc

3. AF-Cluster (Wayment-Steele et al.) — the MSA-clustering method SF-Cluster is
   compared against and conceptually extends.
   License: MIT.
   Source:  https://github.com/HWaymentSteele/AF_Cluster

AlphaFold2 / ColabFold are used as the downstream structure predictor and are
neither included nor wrapped by this package; install them separately under
their own licenses.