run_complexfold.py

#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
#
# ComplexFold is based on AlphaFold2.0. Jumper et al. 2021, 10.1038/s41586-021-03819-2
# https://github.com/deepmind/alphafold
#
#
# Version 1.5
#
# Change Notes
#
# v1.5:
#  - run_reported.py should not use vRAM anymore
#  - run_reported.py can output plot for specified model
#  - run_reported.py calculates the dipTM (domain-interaction pTM). It works like the ipTM but it compares specified
#    regions even if they are on the same chain.
#  - Changed the colour scheme of the plots
#  - Introduced the ipTM
#  - Skip amber relaxation
#
# v1.4:
#  - Added run_reporter.py. This script read AF and CF pickled files and evaluates them.
#  - Changed the pickled output so it contains logits and breaks for the calculation of a pTM for a subset of residues
#  - Changed Result and Result_Handler so they also accept pickled results from AF
#
# v1.3:
#  - Custom MSAs
#  - Cap UniRef MSAs
#
# v1.2:
#  - Added stopping of recycling in dependece on Ca-RMS as in ColabFold
#  - Pickle all unrelaxed proteins and all quality scores in parsed_results.pkl; No big array anymore
#  - Made ranking.json to report.json: More output info and all Flags saved
#  - Added the usage of several random_seed. Complexfold will predict for each seed but relaxes only the top 5
#  (pLDDT or TM-score, depending on models used, top 5 is hardcoded atm)
#  - focus_region: Instead of keeping predictions with high global score, keep those with high average pLDDT in that region
#
# v1.1:
#  - Fixed amber removing TER cards and thus crashing pdbfixer
#  - Made the the pickled data optional (default is False)
#  - ranking.json contains pLDDTs and PAEs
#  - Amber may be run on CUDA -> ~50x speed up
#  - Added colabfold.plot_protein

"""Full AlphaFold protein structure prediction script."""
import os
import pickle
import random
import sys
import time
import pathlib
import subprocess
from typing import Dict

from absl import app
from absl import flags
from absl import logging
from alphafold.common import protein
from alphafold.common import residue_constants
from alphafold.data import complex_pipeline
from alphafold.data import templates
from alphafold.model import data
from alphafold.model import config
from alphafold.model import model
from alphafold.model import features
from alphafold.relax import relax
from alphafold import complexfold

import jax
import numpy as np
# Internal import (7716).

flags.DEFINE_list('fasta_paths', None, 'Paths to FASTA files, each containing '
                  'one sequence. Paths should be separated by commas. '
                  'All FASTA paths must have a unique basename as the '
                  'basename is used to name the output directories for '
                  'each prediction.')
flags.DEFINE_string('output_dir', None, 'Path to a directory that will '
                    'store the results.')
flags.DEFINE_string('msa_library_dir', None, 'Path to a directory that contains '
                    'previously made MSAs and found templates.')
flags.DEFINE_list('model_names', None, 'Names of models to use.')
flags.DEFINE_string('data_dir', None, 'Path to directory of supporting data.')
flags.DEFINE_string('jackhmmer_binary_path', '/usr/bin/jackhmmer',
                    'Path to the JackHMMER executable.')
flags.DEFINE_string('hhblits_binary_path', '/usr/bin/hhblits',
                    'Path to the HHblits executable.')
flags.DEFINE_string('hhsearch_binary_path', '/usr/bin/hhsearch',
                    'Path to the HHsearch executable.')
flags.DEFINE_string('kalign_binary_path', '/usr/bin/kalign',
                    'Path to the Kalign executable.')
flags.DEFINE_string('uniref90_database_path', None, 'Path to the Uniref90 '
                    'database for use by JackHMMER.')
flags.DEFINE_string('mgnify_database_path', None, 'Path to the MGnify '
                    'database for use by JackHMMER.')
flags.DEFINE_string('bfd_database_path', None, 'Path to the BFD '
                    'database for use by HHblits.')
flags.DEFINE_string('small_bfd_database_path', None, 'Path to the small '
                    'version of BFD used with the "reduced_dbs" preset.')
flags.DEFINE_string('uniclust30_database_path', None, 'Path to the Uniclust30 '
                    'database for use by HHblits.')
flags.DEFINE_string('pdb70_database_path', None, 'Path to the PDB70 '
                    'database for use by HHsearch.')
flags.DEFINE_string('template_mmcif_dir', None, 'Path to a directory with '
                    'template mmCIF structures, each named <pdb_id>.cif')
flags.DEFINE_string('max_template_date', None, 'Maximum template release date '
                    'to consider. Important if folding historical test sets.')
flags.DEFINE_string('obsolete_pdbs_path', None, 'Path to file containing a '
                    'mapping from obsolete PDB IDs to the PDB IDs of their '
                    'replacements.')
flags.DEFINE_list('random_seeds', [-1], 'The random seed for the data '
                     'pipeline, comma-separated list. By default, this is randomly generated. Note '
                     'that even if this is set, Alphafold may still not be '
                     'deterministic, because processes like GPU inference are '
                     'nondeterministic.')
flags.DEFINE_integer('num_seeds', 1,
                     'Number of random seeds to during processing i.e. how often shall each model be run.', lower_bound=0)
flags.DEFINE_enum('preset', 'full_dbs',
                  ['reduced_dbs', 'full_dbs', 'casp14'],
                  'Choose preset model configuration - no ensembling and '
                  'smaller genetic database config (reduced_dbs), no '
                  'ensembling and full genetic database config  (full_dbs) or '
                  'full genetic database config and 8 model ensemblings '
                  '(casp14).')
flags.DEFINE_boolean('benchmark', False, 'Run multiple JAX model evaluations '
                     'to obtain a timing that excludes the compilation time, '
                     'which should be more indicative of the time required for '
                     'inferencing many proteins.')
flags.DEFINE_boolean('write_features_models', False, 'Write features.pkl and model resut pkls.')
flags.DEFINE_enum('amber_accel', 'CPU', ['CPU', 'CUDA'], 'Hardware used for Amber refinement.')
flags.DEFINE_boolean('relax', True, 'Shall the output be relaxed with amber.')
flags.DEFINE_integer('num_recycle', 10, 'Number of recycling during prediction.', lower_bound=0)
flags.DEFINE_float('recycling_tolerance', 0.25, 'Tolerance for deciding when to stop recycling (Ca-RMS).', lower_bound=0)
flags.DEFINE_list('focus_region', [],
                  'Focus on position x through y while deciding which result to keep. Uses the mean pLDDT of that '
                  'region. For complexes, concatenate seqcuences and count from the very beginning.')

FLAGS = flags.FLAGS
                    
MAX_TEMPLATE_HITS = 20
RELAX_MAX_ITERATIONS = 0
RELAX_ENERGY_TOLERANCE = 2.39
RELAX_STIFFNESS = 10.0
RELAX_EXCLUDE_RESIDUES = []
RELAX_MAX_OUTER_ITERATIONS = 20


def _check_flag(flag_name: str,
                other_flag_name: str,
                should_be_set: bool):
  if should_be_set != bool(FLAGS[flag_name].value):
    verb = 'be' if should_be_set else 'not be'
    raise ValueError(f'{flag_name} must {verb} set when running with '
                     f'"--{other_flag_name}={FLAGS[other_flag_name].value}".')


def predict_structure(
    fasta_path: str,
    fasta_name: str,
    output_dir_base: str,
    data_pipeline: complex_pipeline.DataPipeline,
    model_runners: Dict[str, model.RunModel],
    amber_relaxer: relax.AmberRelaxation,
    benchmark: bool,
    random_seeds: list,
    arguments_str: str,
    focus_region: list = [],
    write_features_models: bool = False):

  """Predicts structure using AlphaFold for the given sequence."""
  timings = {}
  t_00 = time.time()

  output_dir = os.path.join(output_dir_base, fasta_name)
  if not os.path.exists(output_dir):
    os.makedirs(output_dir)
  msa_output_dir = os.path.join(output_dir, 'msas')
  heteromer_output_dir = os.path.join(output_dir, 'heteromers')

  if not os.path.exists(heteromer_output_dir):
    os.makedirs(heteromer_output_dir)

  if not os.path.exists(msa_output_dir):
    os.makedirs(msa_output_dir)

  if 'msa_coverage.png' in os.listdir(output_dir):
    result_counter = 0
    for os_dir in os.listdir(output_dir):
      if os_dir.startswith('result_'):
        dir = os.path.join(output_dir, os_dir)
        if os.path.isdir(dir):
          result_counter = max(result_counter, int(os_dir.split('_')[1])+1)
    new_dir = os.path.join(output_dir, f'result_{result_counter}')
    os.makedirs(new_dir)
    for os_file in os.listdir(output_dir):
      file_path = os.path.join(output_dir, os_file)
      if os.path.isfile(file_path):
        new_file_path = os.path.join(new_dir, os_file)
        os.rename(file_path, new_file_path)

  # Get features.
  t_0 = time.time()
  feature_dict = data_pipeline.process(
      input_fasta_path=fasta_path,
      output_dir=output_dir,
      msa_output_dir=msa_output_dir,
      heteromer_output_dir=heteromer_output_dir)
  timings['Features'] = time.time() - t_0

  if write_features_models:
    # Write out features as a pickled dictionary.
    features_output_path = os.path.join(output_dir, 'features.pkl')
    with open(features_output_path, 'wb') as f:
      pickle.dump(feature_dict, f, protocol=4)

  result_handler = complexfold.Result_Handler(feature_dict=feature_dict,
                                              random_seeds=random_seeds,
                                              num_results=5,
                                              focus_region=focus_region,
                                              is_ptm=list(model_runners.keys())[0].endswith('ptm'))

  # Run the models.
  for model_name_base, model_runner in model_runners.items():
    for i, random_seed in enumerate(random_seeds):
      model_name = f'{model_name_base}-s{i}'
      logging.info(f'Running {model_name}')

      t_0 = time.time()
      processed_feature_dict = model_runner.process_features(feature_dict, random_seed=random_seed)

      num_templates = 0
      if 'template_mask' in processed_feature_dict:
        num_templates = np.sum(processed_feature_dict['template_mask'].mean(axis=0) > 0)
      logging.info(f'Using {num_templates if num_templates > 0 else "NO"} templates')

      prediction_result = model_runner.predict(processed_feature_dict, random_seed=random_seed)
      # Parse results for ColabFold compatibility
      t_diff = time.time() - t_0
      result_handler.add(complexfold.Result(model_name=model_name,
                                            random_seed=random_seed,
                                            prediction_result=prediction_result,
                                            processed_feature_dict=processed_feature_dict,
                                            Ls=result_handler.Ls,
                                            timing=t_diff))
      timings[f'Process, predict, compile - {model_name}'] = t_diff
      logging.info('Total %s predict time (+ compilation): %.0f', model_name, t_diff)

      if benchmark:
        t_0 = time.time()
        model_runner.predict(processed_feature_dict)
        timings[f'Predict benchmark {model_name}'] = time.time() - t_0


  logging.info(f'Relax and report best {result_handler.num_results} models.')
  for result in result_handler.results:
    logging.info('Processing %s', result.name)
    # Save the model outputs.
    if write_features_models:
      result_output_path = os.path.join(output_dir, f'result_{result.name}.pkl')
      with open(result_output_path, 'wb') as f:
        pickle.dump(result, f, protocol=4)
      
    # Save unrelaxed model
    unrelaxed_pdb_path = os.path.join(output_dir, f'unrelaxed_{result.name}.pdb')
    with open(unrelaxed_pdb_path, 'w') as f:
      f.write(
        protein.to_pdb(
          result.unrelaxed_protein)
      )
    
    # Relax the prediction.
    if amber_relaxer is not None:
      t_0 = time.time()
      relaxed_pdb_str, _, _ = amber_relaxer.process(prot=result.unrelaxed_protein)
      timings[f'Relax {result.name}'] = time.time() - t_0

      # Save the relaxed PDB.
      relaxed_output_path = os.path.join(output_dir, f'relaxed_{result.name}.pdb')
      with open(relaxed_output_path, 'w') as f:
        f.write(relaxed_pdb_str)

  # Pickle result handler
  result_handler.pickle(output_dir=output_dir)

  # Plot data
  t_0 = time.time()
  result_handler.plot(output_dir=output_dir, model_names=result_handler.get_model_names())
  timings['Plotting'] = time.time() - t_0
  timings['Total [s]'] = time.time() - t_00
                                
  # Report
  result_handler.report(timings=timings, arguments_str=arguments_str, output_dir=output_dir)

      
def main(argv):
  if len(argv) > 1:
    raise app.UsageError('Too many command-line arguments.')
      
  use_small_bfd = FLAGS.preset == 'reduced_dbs'
  _check_flag('small_bfd_database_path', 'preset',
              should_be_set=use_small_bfd)
  _check_flag('bfd_database_path', 'preset',
              should_be_set=not use_small_bfd)
  _check_flag('uniclust30_database_path', 'preset',
              should_be_set=not use_small_bfd)

  if FLAGS.preset in ('reduced_dbs', 'full_dbs'):
    num_ensemble = 1
  elif FLAGS.preset == 'casp14':
    num_ensemble = 8
        
  model_names = FLAGS.model_names

  # Check for duplicate FASTA file names.
  fasta_names = [pathlib.Path(p).stem for p in FLAGS.fasta_paths]
  if len(fasta_names) != len(set(fasta_names)):
    raise ValueError('All FASTA paths must have a unique basename.')
  elif len(fasta_names) > 1:
    raise ValueError('Use only one FASTA at a time with ComplexFold.')
    
  peptide_docking = False
  with open(FLAGS.fasta_paths[0], 'r') as f:
    heteromers = set()
    num_heteromers = 0
    for line in f.readlines():    
      if line.startswith('>'):
        if not line.strip() in heteromers:
          heteromers.add(line.strip())
          num_heteromers += 1
    
        if line.rstrip() == '>Peptide':
          peptide_docking = True

  if num_heteromers == 1:
    global MAX_TEMPLATE_HITS
    logging.info('%d entries in FASTA. -> Normal AlphaFold', num_heteromers)

  elif num_heteromers > 1:
    model_names = []
    for model_name in FLAGS.model_names:
      if not model_name.endswith('ptm'):
        model_names.append(model_name + '_ptm')
      else:
        model_names.append(model_name)

    if not peptide_docking:
      MAX_TEMPLATE_HITS = 10
      logging.info('%d entries in FASTA. -> Switch to complex mode:', num_heteromers)
      logging.info(' - Set MAX_TEMPLATE_HITS per heteromer to %d', MAX_TEMPLATE_HITS)
      logging.info(' - Multiplied max_templates by number of heteromers (%d)', num_heteromers)
    else:
      logging.info('Peptide found. -> Switch to peptide docking mode:')
              
    logging.info(' - Enforce ptm models')
              
  template_featurizer = templates.TemplateHitFeaturizer(
      mmcif_dir=FLAGS.template_mmcif_dir,
      max_template_date=FLAGS.max_template_date,
      max_hits=MAX_TEMPLATE_HITS,
      kalign_binary_path=FLAGS.kalign_binary_path,
      release_dates_path=None,
      obsolete_pdbs_path=FLAGS.obsolete_pdbs_path)
        
  data_pipeline = complex_pipeline.DataPipeline(
      jackhmmer_binary_path=FLAGS.jackhmmer_binary_path,
      hhblits_binary_path=FLAGS.hhblits_binary_path,
      hhsearch_binary_path=FLAGS.hhsearch_binary_path,
      uniref90_database_path=FLAGS.uniref90_database_path,
      mgnify_database_path=FLAGS.mgnify_database_path,
      bfd_database_path=FLAGS.bfd_database_path,
      uniclust30_database_path=FLAGS.uniclust30_database_path,
      small_bfd_database_path=FLAGS.small_bfd_database_path,
      pdb70_database_path=FLAGS.pdb70_database_path,
      msa_library_dir=FLAGS.msa_library_dir,
      template_featurizer=template_featurizer,
      use_small_bfd=use_small_bfd)

      
  model_runners = {}
  for model_name in model_names:
    model_config = config.model_config(model_name)
    model_config.data.eval.num_ensemble = num_ensemble
    model_config.data.eval.max_templates = num_heteromers * model_config.data.eval.max_templates
    model_config.model.embeddings_and_evoformer.template.max_templates = model_config.data.eval.max_templates
    model_config.data.common.num_recycle = FLAGS.num_recycle
    model_config.model.num_recycle = FLAGS.num_recycle
    model_config.model.recycle_tol = FLAGS.recycling_tolerance
    model_params = data.get_model_haiku_params(model_name=model_name, data_dir=FLAGS.data_dir)
    model_runner = model.RunModel(model_config, model_params)
    model_runners[model_name] = model_runner
    
  logging.info('Have %d models: %s', len(model_runners),
               list(model_runners.keys()))

  amber_relaxer = None
  if FLAGS.relax:
    amber_relaxer = relax.AmberRelaxation(
        max_iterations=RELAX_MAX_ITERATIONS,
        tolerance=RELAX_ENERGY_TOLERANCE,
        stiffness=RELAX_STIFFNESS,
        exclude_residues=RELAX_EXCLUDE_RESIDUES,
        max_outer_iterations=RELAX_MAX_OUTER_ITERATIONS,
        platform_name=FLAGS.amber_accel
    )
    
  random_seeds = [int(i) for i in FLAGS.random_seeds]
  if random_seeds == [-1]:
    random_seeds = [random.randrange(sys.maxsize) for _ in range(FLAGS.num_seeds)]
  logging.info(f'Using random seeds for the data pipeline: {random_seeds}')
  
  # Predict structure for each of the sequences.
  for fasta_path, fasta_name in zip(FLAGS.fasta_paths, fasta_names):
    predict_structure(
        fasta_path=fasta_path,
        fasta_name=fasta_name,
        output_dir_base=FLAGS.output_dir,
        data_pipeline=data_pipeline,
        model_runners=model_runners,
        amber_relaxer=amber_relaxer,
        benchmark=FLAGS.benchmark,
        random_seeds=random_seeds,
        write_features_models=FLAGS.write_features_models,
        focus_region=[int(i) for i in FLAGS.focus_region],
        arguments_str=FLAGS.flags_into_string())
      
    
if __name__ == '__main__':
  flags.mark_flags_as_required([
      'fasta_paths',
      'output_dir',
      'model_names',
      'data_dir',
      'uniref90_database_path',
      'mgnify_database_path',
      'pdb70_database_path',
      'template_mmcif_dir',
      'max_template_date',
      'obsolete_pdbs_path',
  ])
    
  flags.register_validator('model_names',
                         checker=lambda liste: all([model_name.endswith('ptm') for model_name in liste]) or \
                                               not any([model_name.endswith('ptm') for model_name in liste]),
                         message='All models must either be normal or "_ptm".')
  flags.register_validator('random_seeds',
                         checker=lambda liste: all([complexfold.is_int(i) for i in liste]),
                         message='Random seeds must be integers.')
  flags.register_validator('focus_region',
                         checker=lambda liste: all([complexfold.is_int(i) for i in liste]) and (len(liste) == 2 or len(liste) == 0),
                         message='Focus region must be defined by two comma-separated integers.')

  app.run(main)