Getting Started

Steps to set up a local installation of our modified version of AlphaFold2.

Prerequisites

Installation of Anaconda is required to load dependencies.

• Installing conda: conda-install-link

Installation

1. Clone the repo:

   git clone https://github.com/Kuhlman-Lab/alphafold.git

2. Load AlphaFold2 model weights from source using script: https://github.com/Kuhlman-Lab/alphafold/blob/main/setup/download_alphafold_params.sh

3. Set up conda environment:

   conda env create -n alphafold -f setup/af2_env.yml
   conda activate alphafold
   pip3 install --upgrade jax==0.3.25 jaxlib==0.3.25+cuda11.cudnn805 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
   python3 -m pip install /path/to/alphafold/alphafold/

Usage

The input file should be a CSV file. Each new line in the CSV will be identified as a separate query and will generate an individual prediction. Each line should contain a sequence starting with a comma, and each chain is separated by a comma.

Find below the flags that can be provided to the model runner, grouped by functionality:

Input/Output

Flag	Type	Default	Description
--input_dir	str	''	Directory containing input files.
--output_dir	str	''	Directory to store prediction results. Default is ./prediction_{datetime}.
--params_dir	str	''	Path to parameter weights directory. Default is ../alphafold/data/.
--compress_output	flag	False	Compress result dictionaries per query.
--dont_write_pdbs	flag	False	Skip writing PDB files.
--save_timing	flag	False	Save runtime metrics as a .pkl file.
--batch_predictions	flag	False	Enable batched predictions.
--no_logging	flag	False	Disable logging output.
--design_run	flag	False	Suppress most outputs unless overridden.
--initial_guess	str	None	Path to initial guess PDB file.

Sequence control

Flag	Type	Default	Description
--min_length	int	1	Minimum chain length for prediction.
--max_pad_size	int	None	Pad short sequences to this length to avoid recompilation.
--max_length	int	2500	Maximum single-chain length.
--max_multimer_length	int	2500	Maximum total multimer length. Note: validation limited above 1536 residues.

MSA generation

Flag	Type	Default	Description
--msa_mode	str	'MMseqs2-U+E'	MSA mode. Options: MMseqs2-U+E, MMseqs2-U, single_sequence.
--custom_msa_path	str	None	Path to custom .a3m files. msa_mode should be single_sequence to use this flag.
--insert_msa_gaps	flag	False	Add gaps to custom MSAs based on sequence differences.
--update_msa_query_seq	float	1.0	Sequence identity threshold for updating custom MSAs.

Template usage

Flag	Type	Default	Description
--use_templates	flag	False	Use templates from MMseqs2.
--max_template_date	str	'2100-01-01'	Max release date for templates (YYYY-MM-DD).
--custom_template_path	str	None	Path to custom template PDBs.
--rm_template_seq	flag	False	Remove sequences from templates.
--dont_mask_template_interchain	flag	False	Enable interchain template usage.
--permute_templates	flag	False	Enable symmetric chain template permutation.

Model control

Flag	Type	Default	Description
--use_ptm	flag	False	Use pTM models for predicted aligned error (PAE).
--no_multimer_models	flag	False	Use monomer models even for multimers.
--use_multimer_v1	flag	False	Use original v1 multimer weights.
--use_multimer_v2	flag	False	Use v2 multimer weights (instead of v3).
--num_models	int	5	Number of models to run (1–5).
--num_ensemble	int	1	Number of MSA ensemble runs per model (CASP14 used 8).
--random_seed	int	None	Random seed for reproducibility.
--num_seeds	int	1	Number of seeds per model/sequence.
--is_training	flag	False	Enable dropout to sample diverse structures.
--max_recycle	int	3	Max recycling steps for refinement.
--recycle_tol	float	0.0	Early stop recycling if Ca-RMS change < tol.

Other options

Flag	Type	Default	Description
--use_amber	flag	False	Run Amber relaxation (improves geometry but increases runtime).
--__override_disable	flag	False	Internal override for debugging/disabling features.

(back to top)

Running and Analysis

To make a prediction, provide the sequences in the input CSV file. Set the options in the af2.flags file. Then, run the commands:

conda activate alphafold
python /path/to/alphafold/run/run_af2.py @af2.flags

Running the code will produce an outputs directory containing PDB files with predicted structures and .pbz2 files with compressed information about the predictions such as confidence metrics. To generate pLDDT plots and PAE heatmaps, navigate to a directory containing .pbz2 files and run the command:

conda activate alphafold
python /path/to/alphafold/run/plots.py

Specific examples

Specify your AF2 options in af2.flags.

To turn off MSA generation:

--msa_mode single_sequence  

To use precomputed MSA:

--msa_mode single_sequence  
--custom_msa_path /path/to/directory/with/a3m/files/

To use custom templates:

--use_templates  
--custom_template_path templates

Since we have modified the original AF2 code to allow for custom template databases, you will have to make sure each .pdb file within the templates folder has a name consisting of 4 letters and numbers, essentially mimicking a file from the PDB database with a PDB code (although the file name does not actually have to be a real PDB code). Some examples could be “temp.pdb”, “1uzg.pdb”, “PPPP.pdb”, etc.

(back to top)