usage.md

Run with Command line

If you set up IntelliFold by pip, you can run the following command to do model inference:

# run with example yaml, which contains precomputed msa files(.a3m or .csv). the default seed is 42.
## a3m MSA file type
intellifold predict ./examples/5S8I_A.yaml --out_dir ./output 
## csv MSA file type
intellifold predict ./examples/7yds.yaml --out_dir ./output 

# Predict with a directory of YAMLs
intellifold predict ./examples --out_dir ./output

# run with 5 seeds(mutiple seeds are splited them by comma) and 5 samples (the default parameters for AlphaFold3).
intellifold predict ./examples/5S8I_A.yaml --seed 42,43,44,45,46 --num_diffusion_samples 5 --out_dir ./output

# if the input yaml file do not contain precomputed msa paths, you can set --use_msa_server to search msa(need internet connection and would take some time) 
# and use greedy msa pairing strategy
# and then predict
intellifold predict examples/examples_wo_msa/example_without_msa.yaml --out_dir ./output --seed 42,66 --use_msa_server --msa_pairing_strategy greedy

# only run the data processing step, and not run the model.
intellifold predict ./examples/5S8I_A.yaml --out_dir ./output --only_run_data_process

# choose the model version to run, the default is v2-flash, which is faster and more accurate than v1 and v2. you can choose the model based on your needs and computational resources.
intellifold predict ./examples/5S8I_A.yaml --out_dir ./output --model v2

# predict with use template online search
intellifold predict ./examples/examples_wo_msa/example_without_msa.yaml --out_dir ./output --use_template --use_msa_server

# predict with precompute protein msa and templates
intellifold predict ./examples/5S8I_A_wtemplate.yaml --out_dir ./output --use_template

Run with Bash Script

The aurguments is the same as intellifold predict, and you can set the parameters in the script. you can get the help information by running intellifold predict --help or python run_intellifold.py --help

bash predict.sh
## or
## python run_intellifold.py ....

Common arguments of this scripts/intellifold predict are explained as follows:

• --out_dir (PATH, default: ./)
  The path where to save the predictions.
• --cache (PATH, default: ~/.intellifold)
  The directory where to download the data and model. Will use environment variable INTELLIFOLD_CACHE as an absolute path if set.
• --num_workers (INTEGER, default: 4)
  The number of dataloader workers to use for prediction.
• --precision (str, default: bf16)
  Sets precision, lower precision improves runtime.
• --seed (INTEGER, default: 42)
  Random seed (single int or multiple ints separated by comma, e.g., '42' or '42,43').
• --recycling_iters (INTEGER, default: 10)
  Number of recycling iterations.
• --num_diffusion_samples (INTEGER, default: 5)
  The number of diffusion samples.
• --sampling_steps (INTEGER, default: 200)
  The number of diffusion sampling steps to use.
• --output_format ([pdb,mmcif], default: mmcif)
  The output format to use for the predictions (pdb or mmcif).
• --override (FLAG, default: False)
  Whether to override existing found predictions.
• --use_msa_server (FLAG, default: False)
  Whether to use the MMSeqs2 server for MSA generation.
• --msa_server_url (str, default: https://api.colabfold.com)
  MSA server url. Used only if --use_msa_server is set.
• --msa_pairing_strategy (str, default: complete)
  Pairing strategy to use. Used only if --use_msa_server is set. Options are 'greedy' and 'complete'.
• --no_pairing (FLAG, default: False)
  Whether to use pairing for Protein Multimer MSA generation.
• --use_template (FLAG, default: False)
  Whether to use template information for prediction. If set, the model will use the template features in the input YAML file (if provided) or search for templates online (if --use_msa_server is set).
• --only_run_data_process (FLAG, default: False)
  Whether to only run data processing, and not run the model.
• --return_similar_seq (FLAG, default: False) Whether to return sequences similar to those in the training PDB dataset during inference. You can use these similar sequences and its PDB ids to do further analysis, such as a reference structure.

  Before using this option, please make sure the mmseqs2 tool is installed, you can install it by running conda install -c conda-forge -c bioconda mmseqs2

• --model ([v1, v2, v2-flash], default: v2-flash)
  The model to use for prediction. Options are 'v1', 'v2', and 'v2-flash'. 'v2-flash' is the default and recommended model, which is faster and more accurate than 'v1' and 'v2'. 'v1' is the original model used in the IntelliFold paper, and 'v2' is an improved version of the model with better performance but slower inference speed than 'v2-flash'. You can choose the model based on your needs and computational resources.

Tools for Generating the Template

IntelliFold provides a script run_templates_search.py for searching templates. We recommend using an MSA generated against UniRef (e.g., UniRef90/UniRef100) as the input for template search for better coverage and more reliable hits. You can run the script as follows:

## example command for searching templates
python runner/run_templates_search.py \
--input_msa ./examples/msas/5s8i_A.a3m  \
--output_template ./output/5s8i_A_hmmsearch.a3m \
--seqres_database_path /path/to/pdb_seqres_2022_09_28.fasta

common arguments of run_templates_search.py are explained as follows:

• --input_msa (PATH, required)
  Path of containing MSA a3m/csv file for template search.
• --output_template (PATH, required)
  Path to save the hmmsearch a3m result, templates result.
• --hmmsearch_binary_path (PATH, default: None)
  Path to hmmsearch binary. If not provided, will try to find it in the system PATH.
• --hmmbuild_binary_path (PATH, default: None)
  Path to hmmbuild binary. If not provided, will try to find it in the system PATH.
• --seqres_database_path (PATH, default: None)
  Path to sequence database. If not provided, will use the default pdb_seqres_2022_09_28.fasta database in the intellifold cache directory, if the database file is not found in the cache directory, it will be downloaded automatically, but you need to set the INTELLIFOLD_CACHE environment variable.