from dataclasses import dataclass, field from pathlib import Path from modeller import * from modeller.automodel import * import time from typing import List, Tuple import sys import glob import pyfastx from Bio import SeqIO from Bio.Seq import Seq from Bio.SeqRecord import SeqRecord @dataclass class PDBModeler: structure_file: Path fasta_file: Path outdir: Path chain: str num_loop: int = 2 md_level: str = 'refine.fast' # refine.very_fast or refine.slow optional def __post_init__(self): self.structure = self.structure_file.stem self.ali_file = self.fasta_to_ali() @staticmethod def find_non_dash_indices(seq): start = next((i for i, c in enumerate(seq) if c != '-'), None) end = next((i for i, c in enumerate(reversed(seq)) if c != '-'), None) if start is not None and end is not None: end = len(seq) - end return start, end @staticmethod def align_sequences(file: Path) -> Path: fx = pyfastx.Fasta(file.as_posix(), build_index=True) assert len(fx) == 2 seqs = [seq for seq in fx] # 确定哪条链需要裁剪 if seqs[0].seq.startswith('-') or seqs[0].seq.endswith('-'): trim_index = 0 elif seqs[1].seq.startswith('-') or seqs[1].seq.endswith('-'): trim_index = 1 else: # 如果两条链都不需要裁剪，就直接返回原文件 return file start, end = PDBModeler.find_non_dash_indices(seqs[trim_index].seq) # 根据确定的裁剪位置裁剪两条链 trimmed_seqs = [] for seq in seqs: trimmed_seq = seq.seq[start:end] trimmed_seqs.append(SeqRecord(Seq(trimmed_seq), id=seq.name, description="")) # 选择没有'-'的序列 selected_seq = None for seq_record in trimmed_seqs: if '-' not in seq_record.seq: selected_seq = seq_record break assert selected_seq is not None, "no sequence without '-' found" assert not selected_seq.seq.startswith('-') and not selected_seq.seq.endswith('-'), "selected sequence should not start or end with '-'" # Write the selected sequence to a new FASTA file using Biopython new_fasta_file = file.with_suffix('.selected.fasta') with open(new_fasta_file, 'w') as output_handle: SeqIO.write([selected_seq], output_handle, "fasta") return new_fasta_file def _make_model_env(self, ali_file: Path) -> Tuple[Path, Path]: ''' ali_file: 需要和PDB文件中对齐的 ali文件，这里在类初始化的时候就将对应的fasta文件转化了ali文件格式（self.fasta_to_ali()），这里是对齐操作 ''' env = Environ() mdl = Model(env, file=self.structure_file.as_posix(), model_segment=(f'FIRST:{self.chain}', f'LAST:{self.chain}')) aln = Alignment(env) aln.append_model(mdl, align_codes=self.structure, atom_files=self.structure_file.as_posix()) aln.append(file=ali_file.as_posix(), align_codes=self.structure) aln.align2d() new_ali_file = self.outdir / f'alignment_{self.chain}.ali' aln.write(file=new_ali_file.as_posix(), alignment_format='PIR') aln.write(file=(self.outdir / f'alignment_{self.chain}.pap').as_posix(), alignment_format='PAP') fasta_ali_file = self.outdir / f'alignment_{self.chain}.fasta' aln.write(file=fasta_ali_file.as_posix(), alignment_format='FASTA') return fasta_ali_file, new_ali_file def make_model(self): # 单模板建模 print("***************************************************") print("md_level ====", self.md_level) print("***************************************************") time_start = time.time() # env1 = Environ() # mdl = Model(env1, file=self.structure_file.as_posix(), model_segment=(f'FIRST:{self.chain}', f'LAST:{self.chain}')) # aln = Alignment(env1) # aln.append_model(mdl, align_codes=f'{self.structure}_{self.chain}', atom_files=self.structure_file.as_posix()) # aln.append(file=self.ali_file.as_posix(), align_codes=self.fasta_file.stem) # aln.align2d() # aln.write(file=(self.outdir / 'alignment1.ali').as_posix(), alignment_format='PIR') # aln.write(file=(self.outdir / 'alignment1.pap').as_posix(), alignment_format='PAP') # save alignment in FASTA format # aln.write(file=(self.outdir / 'alignment1.fasta').as_posix(), alignment_format='FASTA') fasta_align1, align1 = self._make_model_env(self.ali_file) # 第一次对齐 slice_fasta = PDBModeler.align_sequences(fasta_align1) # 首尾部对齐剔除 fx = pyfastx.Fasta(slice_fasta.as_posix(), build_index=True) if len(fx) == 2 and (slice_fasta == fasta_align1): # 针对首位都没有'-'的序列处理，直接使用'alignment1.fasta'进行同源建模 fix_ali_file = slice_fasta elif len(fx) == 1: slice_ali = self.outdir / 'alignment_slice.ali' PDBModeler.write_ali(slice_ali, f'{fx[0].name}', fx[0].seq) fasta_align2, align2 = self._make_model_env(slice_ali) # 第2次对齐 # env2 = Environ() # mdl2 = Model(env2, file=self.structure_file.as_posix(), model_segment=(f'FIRST:{self.chain}', f'LAST:{self.chain}')) # aln2 = Alignment(env2) # aln2.append_model(mdl2, align_codes=f'{self.structure}_{self.chain}', atom_files=self.structure_file.as_posix()) # aln2.append(file=slice_ali.as_posix(), align_codes=self.fasta_file.stem) # aln2.align2d() # aln2.write(file=(self.outdir / 'alignment2.ali').as_posix(), alignment_format='PIR') # aln2.write(file=(self.outdir / 'alignment2.pap').as_posix(), alignment_format='PAP') # choose ali file fix_ali_file = align2 else: raise Exception(f"Unexpected FASTA file {slice_fasta.as_posix()} in errror number more than 2.") env = Environ() env.io.atom_files_directory = ['.'] loop_model = LoopModel(env, alnfile=fix_ali_file.as_posix(), knowns=self.structure, sequence=self.fasta_file.stem, loop_assess_methods=(assess.DOPE, assess.GA341)) # 设置循环模型数量 # 数量规则：(end - start) + 1 loop_model.loop.starting_model = 1 loop_model.loop.ending_model = self.num_loop # 设置 MD 优化函数为 "refine.slow" 或 "refine.fast if self.md_level.strip() == 'refine.slow': loop_model.loop.md_level = refine.slow elif self.md_level.strip() == 'refine.very_fast': loop_model.loop.md_level = refine.very_fast elif self.md_level.strip() == 'refine.fast': loop_model.loop.md_level = refine.fast # 调用 LoopModel 的 make 方法 loop_model.make() end_time = time.time() print(f"Time cost: {end_time - time_start}s") # 获取所有成功生成的模型文件的路径 model_files = self.get_model_files(loop_model) if model_files: print(f"Model files: {[file.name for file in model_files]}") else: print("No model files found.") return model_files def find_pdb95_fsa_file(self) -> Path: """在 Conda 环境中查找 pdb95.fsa 文件的路径。""" # 获取当前 Python 解释器的路径 python_executable_path = Path(sys.executable) # 获取 Conda 环境的根目录 conda_env_root = python_executable_path.parent.parent # 获取可能的 Modeller 目录 modeller_dirs = list(conda_env_root.glob("lib/modeller-*/examples/commands")) # 选择最新版本的 Modeller 目录 modeller_dirs.sort(reverse=True) if modeller_dirs: latest_modeller_dir = modeller_dirs[0] pdb95_fsa_path = latest_modeller_dir / "pdb95.fsa" return pdb95_fsa_path else: raise FileNotFoundError("Modeller directory not found.") def get_model_files(self, loop_model) -> List[Path]: # 检查 loop_model.loop.outputs 列表，收集所有成功生成的模型文件 model_files = [] for output in loop_model.loop.outputs: if output.get('failure') is None: model_files.append(Path(output.get('name'))) return model_files @staticmethod def write_ali(ali_file: Path, description: str, sequence: str): with open(ali_file, 'w') as f: f.write(f'>P1;{description}\n') f.write(f'sequence:{description}:::::::0.00: 0.00\n') f.write(f'{sequence}*') def fasta_to_ali(self) -> Path: if not self.outdir.exists(): self.outdir.mkdir(parents=True, exist_ok=True) ali_file = self.outdir / f'{self.fasta_file.stem}.ali' if ali_file.exists(): ali_file.unlink() fx = pyfastx.Fasta(self.fasta_file.as_posix(), build_index=True) assert len(fx) == 1, "FASTA file should contain only one sequence" PDBModeler.write_ali(ali_file, self.structure, fx[0].seq) return ali_file if __name__ == "__main__": import argparse parser = argparse.ArgumentParser(description="Build model by Modeller") parser.add_argument("-s", "--structure", help="Structure file") parser.add_argument("-o", "--outdir", help="Output directory") parser.add_argument("-f", "--fasta", help="Fasta file") parser.add_argument("-n", "--num_loop", help="Number of loop model") parser.add_argument("-m", "--md_level", help="MD level") parser.add_argument("-c", "--chain", help="Chain ID") args = parser.parse_args() modeler = PDBModeler(Path(args.structure), Path(args.fasta), Path(args.outdir), args.chain, int(args.num_loop), args.md_level) modeler.make_model() # test command # python build_modellel.py -s ../5sws_fixer.pdb -o ./5swsmodellerfix -f ../rcsb_pdb_5SWS.fasta -n 1 -m refine.very_fast -c D # python build_modeller.py -s 1j8h.pdb -o ./1j8hmodellerfix -f ./1j8h_D.fasta -n 1 -m refine.very_fast -c D