feat/docs/cases: Covalent Bond Input

[YaoYinYing]

This is one of the separated PRs from #321 .

Full PR roadmap

id	purpose	# commits	Affected
1	Hydra-Omegaconf and Pip module	6	Code, Config, Doc
2	BFD supports and MSA parallelism fixes	5	Code, Config, Case
3	Small molecule inputs and covalent bonds	4	Code, Case, Doc

Changelog

Added

• Support for SDF file inputs for small molecules and additional file formats.
• Expanded set of demo cases.
• PTM Inputs (phosphorylations, isopeptides, metal bindings, disulfide bridges, glycosylations, etc.)

Fixed

• Implemented automatic binary search within the PATH if not explicitly specified.
• Resolved an issue with dummy assertions of ref_atom_name_chars and problematic atoms that caused errors.
• Addressed the None-value in TemplateAtomMaskAllZerosError to prevent incorrect warning triggers.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[alephreish]

@YaoYinYing I had to bring back a couple of imports in order to make it work:

diff --git a/apps/protein_folding/helixfold3/infer_scripts/feature_processing_aa.py b/apps/protein_folding/helixfold3/infer_scripts/feature_processing_aa.py index 603e8da..c994b0d 100644 --- a/apps/protein_folding/helixfold3/infer_scripts/feature_processing_aa.py +++ b/apps/protein_folding/helixfold3/infer_scripts/feature_processing_aa.py @@ -19,6 +19,7 @@ import os from pathlib import Path import pickle from typing import List, Mapping, Optional, Tuple +import time import numpy as np import logging @@ -28,7 +29,7 @@ from helixfold.data import pipeline_multimer from helixfold.data import pipeline_rna_multimer from helixfold.data import pipeline_conf_bonds, pipeline_token_feature, pipeline_hybrid from helixfold.data import label_utils - +from concurrent.futures import ProcessPoolExecutor, as_completed from helixfold.data.tools import utils from .preprocess import Entity, digit2alphabet diff --git a/apps/protein_folding/helixfold3/inference.py b/apps/protein_folding/helixfold3/inference.py index 429809b..9edc0a1 100644 --- a/apps/protein_folding/helixfold3/inference.py +++ b/apps/protein_folding/helixfold3/inference.py @@ -24,6 +24,7 @@ import pickle import pathlib import shutil import numpy as np +import logging from helixfold.common import all_atom_pdb_save from helixfold.data.pipeline_conf_bonds import load_ccd_dict from helixfold.model import config, utils @@ -116,7 +117,7 @@ def resolve_bin_path(cfg_path: str, default_binary_name: str)-> str: raise FileNotFoundError(f"Could not find a proper binary path for {default_binary_name}: {cfg_path}.") -def get_msa_templates_pipeline(cfg: DictConfig) -> Dict: +def get_msa_templates_pipeline(cfg) -> Dict: use_precomputed_msas = True # Assuming this is a constant or should be set globally template_searcher = hmmsearch.Hmmsearch( diff --git a/apps/protein_folding/helixfold3/utils/model.py b/apps/protein_folding/helixfold3/utils/model.py index 4a5b2d6..2ba6337 100644 --- a/apps/protein_folding/helixfold3/utils/model.py +++ b/apps/protein_folding/helixfold3/utils/model.py @@ -17,6 +17,7 @@ import numpy as np import paddle import paddle.nn as nn +import logging import io from helixfold.model import modules_all_atom

Also a side note: leave_atom_flag is currently ignored, so the user has to modify ccd_preprocessed_etkdg.pkl.gz to remove the atoms that leave upon formation of the corresponding covalent bond.

[YaoYinYing]

@alephreish Hi Andrey, this PR is a cherry-pick(which means there exists some poteintal bugs) from one branch that has been definitely out-of-dated. If you are looking for a full-featured branch for you project, please consider this fork.

[alephreish]

@YaoYinYing I've seen the main branch in your fork. I personally like the current interface of helixfold - it's flexible enough, although switching to your new interface would not be a big deal. I have the feeling that this small PR does have a chance of being merged into PaddlePaddle/PaddleHelix:dev since patch-hydra diverged too much by now.