# CASP servers: af2-standard and af2-multimer

This is a pipeline to run vanilla AlphaFold v2.2 on CASP targets.

Databases were downloaded with the scripts provided by DeepMind on April 22nd, 2022.
Uniclust30 was updated to version [2021_03](http://gwdu111.gwdg.de/~compbiol/uniclust/2021_03/).
PDB mmcif files were updated on May, 2nd

The flag file for the monomer pipeline is `monomer_full_dbs.flag`
The flag file for the multimer pipeline is `multimer_full_dbs.flag`

Conda environment export file: `conda_environment.yaml`

05/05: The kalign version that comes with bioconda (3.2.2) crashed on targets T1109 and T1110, so a newer version has been installed from the [source code (3.3.2)](https://github.com/TimoLassmann/kalign/archive/refs/tags/v3.3.2.tar.gz)

13/05: Whenever a multimeric target is too large to be predicted within the server window, we try to model a subset of its chains (e.g. A8 instead of A16, A1B2C2 instead of A2B4C4). Whenever this happens, it is specified in the file naming (e.g. `models_A1B2C2.tar.gz`). `models.tar.gz` is still included in the folder as a symbolic link to keep naming consistent

31/05: Homomeric targets are modelled both as multimers and monomers. Monomers are uploaded in folders with naming `TXXXX_A1`

OS: Red Hat Enterprise Linux release 8.5 (Ootpa)

AlphaFold invocation:

```
export TF_FORCE_UNIFIED_MEMORY=1
export XLA_PYTHON_CLIENT_MEM_FRACTION={n_gpus}
python {alphafold_path} --flagfile {flagfile} --output_dir {results_dir} --fasta_paths {input_fasta}
```