servalcat 0.4.99__cp310-cp310-macosx_10_14_x86_64.whl

servalcat/refine/refine_xtal.py

@@ -0,0 +1,281 @@
+"""
+Author: "Keitaro Yamashita, Garib N. Murshudov"
+MRC Laboratory of Molecular Biology
+    
+This software is released under the
+Mozilla Public License, version 2.0; see LICENSE.
+"""
+from __future__ import absolute_import, division, print_function, generators
+import gemmi
+import numpy
+import pandas
+import os
+import shutil
+import argparse
+from servalcat.utils import logger
+from servalcat import utils
+from servalcat.xtal.sigmaa import decide_mtz_labels, process_input, calculate_maps, calculate_maps_int, calculate_maps_twin
+from servalcat.refine.xtal import LL_Xtal
+from servalcat.refine.refine import Geom, Refine, update_meta, print_h_options
+from servalcat.refmac import refmac_keywords
+from servalcat import ext
+b_to_u = utils.model.b_to_u
+
+def add_arguments(parser):
+    parser.description = "program to refine crystallographic structures"
+    parser.add_argument("--hklin", required=True)
+    parser.add_argument("-d", '--d_min', type=float)
+    parser.add_argument('--d_max', type=float)
+    parser.add_argument('--nbins', type=int, 
+                        help="Number of bins (default: auto)")
+    parser.add_argument("--labin", help="F,SIGF,FREE input")
+    parser.add_argument('--free', type=int,
+                        help='flag number for test set')
+    parser.add_argument('--model', required=True,
+                        help='Input atomic model file')
+    parser.add_argument("--monlib",
+                        help="Monomer library path. Default: $CLIBD_MON")
+    parser.add_argument('--ligand', nargs="*", action="append",
+                        help="restraint dictionary cif file(s)")
+    parser.add_argument('--newligand_continue', action='store_true',
+                        help="Make ad-hoc restraints for unknown ligands (not recommended)")
+    parser.add_argument('--hydrogen', default="all", choices=["all", "yes", "no"],
+                        help="all: add riding hydrogen atoms, yes: use hydrogen atoms if present, no: remove hydrogen atoms in input. "
+                        "Default: %(default)s")
+    parser.add_argument('--hout', action='store_true', help="write hydrogen atoms in the output model")
+    parser.add_argument('--jellybody', action='store_true',
+                        help="Use jelly body restraints")
+    parser.add_argument('--jellybody_params', nargs=2, type=float,
+                        metavar=("sigma", "dmax"), default=[0.01, 4.2],
+                        help="Jelly body sigma and dmax (default: %(default)s)")
+    parser.add_argument('--jellyonly', action='store_true',
+                        help="Jelly body only (experimental, may not be useful)")
+    parser.add_argument('--find_links', action='store_true', 
+                        help='Automatically add links')
+    parser.add_argument('--keywords', nargs='+', action="append",
+                        help="refmac keyword(s)")
+    parser.add_argument('--keyword_file', nargs='+', action="append",
+                        help="refmac keyword file(s)")
+    parser.add_argument('--randomize', type=float, default=0,
+                        help='Shake coordinates with specified rmsd')
+    parser.add_argument('--ncycle', type=int, default=10,
+                        help="number of CG cycles (default: %(default)d)")
+    parser.add_argument('--weight', type=float,
+                        help="refinement weight (default: auto)")
+    parser.add_argument('--no_weight_adjust', action='store_true', 
+                        help='Do not adjust weight during refinement')
+    parser.add_argument('--target_bond_rmsz_range', nargs=2, type=float, default=[0.5, 1.],
+                        help='Bond rmsz range for weight adjustment (default: %(default)s)')
+    parser.add_argument('--ncsr', action='store_true', 
+                        help='Use local NCS restraints')
+    parser.add_argument('--adpr_weight', type=float, default=1.,
+                        help="ADP restraint weight (default: %(default)f)")
+    parser.add_argument('--occr_weight', type=float, default=1.,
+                        help="Occupancy restraint weight (default: %(default)f)")
+    parser.add_argument('--bfactor', type=float,
+                        help="reset all atomic B values to specified value")
+    parser.add_argument('--fix_xyz', action="store_true")
+    parser.add_argument('--adp',  choices=["fix", "iso", "aniso"], default="iso")
+    parser.add_argument('--refine_all_occ', action="store_true")
+    parser.add_argument('--max_dist_for_adp_restraint', type=float, default=4.)
+    parser.add_argument('--adp_restraint_power', type=float)
+    parser.add_argument('--adp_restraint_exp_fac', type=float)
+    parser.add_argument('--adp_restraint_no_long_range', action='store_true')
+    parser.add_argument('--adp_restraint_mode', choices=["diff", "kldiv"], default="kldiv")
+    parser.add_argument('--unrestrained',  action='store_true', help="No positional restraints")
+    parser.add_argument('--refine_h', action="store_true", help="Refine hydrogen (default: restraints only)")
+    parser.add_argument('--twin', action="store_true", help="Turn on twin refinement")
+    parser.add_argument("-s", "--source", choices=["electron", "xray", "neutron"], required=True)
+    parser.add_argument('--no_solvent',  action='store_true',
+                        help="Do not consider bulk solvent contribution")
+    parser.add_argument('--use_work_in_est',  action='store_true',
+                        help="Use work reflections in ML parameter estimates")
+    parser.add_argument('--keep_charges',  action='store_true',
+                        help="Use scattering factor for charged atoms. Use it with care.")
+    parser.add_argument("--keep_entities", action='store_true',
+                        help="Do not override entities")
+    parser.add_argument('--allow_unusual_occupancies', action="store_true", help="Allow negative or more than one occupancies")
+    parser.add_argument('-o','--output_prefix')
+    parser.add_argument("--write_trajectory", action='store_true',
+                        help="Write all output from cycles")
+    parser.add_argument("--vonmises", action='store_true',
+                        help="Experimental: von Mises type restraint for angles")
+# add_arguments()
+
+def parse_args(arg_list):
+    parser = argparse.ArgumentParser()
+    add_arguments(parser)
+    return parser.parse_args(arg_list)
+# parse_args()
+
+def main(args):
+    if args.source == "neutron": assert not args.refine_h # we need deuterium fraction handling in LL
+    if args.ligand: args.ligand = sum(args.ligand, [])
+    if not args.output_prefix:
+        args.output_prefix = utils.fileio.splitext(os.path.basename(args.model))[0] + "_refined"
+
+    keywords = []
+    if args.keywords or args.keyword_file:
+        if args.keywords: keywords = sum(args.keywords, [])
+        if args.keyword_file: keywords.extend(l for f in sum(args.keyword_file, []) for l in open(f))
+    params = refmac_keywords.parse_keywords(keywords)
+    params["write_trajectory"] = args.write_trajectory
+    
+    hklin = args.hklin
+    labin = args.labin
+    if labin is not None:
+        labin = labin.split(",")
+    elif utils.fileio.is_mmhkl_file(hklin):
+        hklin = utils.fileio.read_mmhkl(hklin)
+        labin = decide_mtz_labels(hklin)
+        
+    if labin and len(labin) == 3: # with test flags
+        use_in_target = "work"
+        if args.use_work_in_est:
+            use_in_est = "work"
+            n_per_bin = 100
+        else:
+            use_in_est = "test"
+            n_per_bin = 50
+    else:
+        use_in_est = "all"
+        use_in_target = "all"
+        n_per_bin = 100
+
+    try:
+        hkldata, sts, fc_labs, centric_and_selections, args.free = process_input(hklin=hklin,
+                                                                                 labin=labin,
+                                                                                 n_bins=args.nbins,
+                                                                                 free=args.free,
+                                                                                 xyzins=[args.model],
+                                                                                 source=args.source,
+                                                                                 d_max=args.d_max,
+                                                                                 d_min=args.d_min,
+                                                                                 n_per_bin=n_per_bin,
+                                                                                 use=use_in_est,
+                                                                                 max_bins=30,
+                                                                                 keep_charges=args.keep_charges,
+                                                                                 allow_unusual_occupancies=args.allow_unusual_occupancies)
+    except RuntimeError as e:
+        raise SystemExit("Error: {}".format(e))
+
+    is_int = "I" in hkldata.df
+    st = sts[0]
+    utils.model.fix_deuterium_residues(st)
+    if args.unrestrained:
+        monlib = gemmi.MonLib()
+        topo = None
+        h_change = gemmi.HydrogenChange.NoChange
+        if args.hydrogen == "all":
+            logger.writeln("\nWARNING: in unrestrained refinement hydrogen atoms are not generated.\n")
+        elif args.hydrogen == "no":
+            st.remove_hydrogens()
+            h_change = gemmi.HydrogenChange.Remove
+        for i, cra in enumerate(st[0].all()):
+            cra.atom.serial = i + 1
+    else:
+        try:
+            monlib = utils.restraints.load_monomer_library(st, monomer_dir=args.monlib, cif_files=args.ligand,
+                                                           stop_for_unknowns=not args.newligand_continue,
+                                                           params=params)
+        except RuntimeError as e:
+            raise SystemExit("Error: {}".format(e))
+        if not args.keep_entities:
+            utils.model.setup_entities(st, clear=True, force_subchain_names=True, overwrite_entity_type=True)
+        utils.restraints.find_and_fix_links(st, monlib, find_metal_links=args.find_links,
+                                            add_found=args.find_links)
+        h_change = {"all":gemmi.HydrogenChange.ReAddKnown,
+                    "yes":gemmi.HydrogenChange.NoChange,
+                    "no":gemmi.HydrogenChange.Remove}[args.hydrogen]
+        try:
+            topo, _ = utils.restraints.prepare_topology(st, monlib, h_change=h_change,
+                                                        check_hydrogen=(args.hydrogen=="yes"),
+                                                        params=params)
+        except RuntimeError as e:
+            raise SystemExit("Error: {}".format(e))
+
+    print_h_options(h_change, st[0].has_hydrogen(), args.refine_h, args.hout, geom_only=False)
+    
+    # initialize ADP
+    utils.model.reset_adp(st[0], args.bfactor, args.adp)
+        
+    # auto weight
+    if args.weight is None:
+        logger.writeln("Estimating weight using resolution")
+        reso = hkldata.d_min_max()[0]
+        args.weight = numpy.exp(reso * 0.9104 + 0.2162)
+        logger.writeln(" Will use weight= {:.2f}".format(args.weight))
+        
+    if args.ncsr:
+        ncslist = utils.restraints.prepare_ncs_restraints(st)
+    else:
+        ncslist = False
+    geom = Geom(st, topo, monlib, shake_rms=args.randomize, adpr_w=args.adpr_weight, occr_w=args.occr_weight, params=params,
+                unrestrained=args.unrestrained or args.jellyonly, use_nucleus=(args.source=="neutron"),
+                ncslist=ncslist)
+    geom.geom.angle_von_mises = args.vonmises
+    geom.geom.adpr_max_dist = args.max_dist_for_adp_restraint
+    if args.adp_restraint_power is not None: geom.geom.adpr_d_power = args.adp_restraint_power
+    if args.adp_restraint_exp_fac is not None: geom.geom.adpr_exp_fac = args.adp_restraint_exp_fac
+    if args.adp_restraint_no_long_range: geom.geom.adpr_long_range = False
+    geom.geom.adpr_mode = args.adp_restraint_mode
+    if args.jellybody or args.jellyonly:
+        geom.geom.ridge_sigma, geom.geom.ridge_dmax = args.jellybody_params
+    if args.jellyonly: geom.geom.ridge_exclude_short_dist = False
+
+    ll = LL_Xtal(hkldata, centric_and_selections, args.free, st, monlib, source=args.source,
+                 use_solvent=not args.no_solvent, use_in_est=use_in_est, use_in_target=use_in_target,
+                 twin=args.twin)
+    refiner = Refine(st, geom, ll=ll,
+                     refine_xyz=not args.fix_xyz,
+                     adp_mode=dict(fix=0, iso=1, aniso=2)[args.adp],
+                     refine_h=args.refine_h,
+                     unrestrained=args.unrestrained,
+                     params=params,
+                     refine_occ=args.refine_all_occ)
+
+    stats = refiner.run_cycles(args.ncycle, weight=args.weight,
+                               weight_adjust=not args.no_weight_adjust,
+                               weight_adjust_bond_rmsz_range=args.target_bond_rmsz_range,
+                               stats_json_out=args.output_prefix + "_stats.json")
+    update_meta(st, stats[-1], ll)
+    refiner.st.name = args.output_prefix
+    utils.fileio.write_model(refiner.st, args.output_prefix, pdb=True, cif=True, hout=args.hout)
+    if params["write_trajectory"]:
+        utils.fileio.write_model(refiner.st_traj, args.output_prefix + "_traj", cif=True)
+
+    if ll.twin_data:
+        # replace hkldata
+        hkldata = calculate_maps_twin(ll.hkldata, ll.b_aniso, ll.fc_labs, ll.D_labs, ll.twin_data,
+                                      centric_and_selections, use=use_in_target)
+    elif is_int:
+        calculate_maps_int(ll.hkldata, ll.b_aniso, ll.fc_labs, ll.D_labs, centric_and_selections,
+                           use=use_in_target)
+    else:
+        calculate_maps(ll.hkldata, ll.b_aniso, centric_and_selections, ll.fc_labs, ll.D_labs, args.output_prefix + "_stats.log",
+                       use=use_in_target)
+
+    # Write mtz file
+    if ll.twin_data:
+        labs = ["F_est", "F_exp", "FOM"]
+    elif is_int:
+        labs = ["I", "SIGI", "FOM"]
+    else:
+        labs = ["FP", "SIGFP", "FOM"]
+    labs.extend(["FWT", "DELFWT", "FC"])
+    if "FAN" in hkldata.df:
+        labs.append("FAN")
+    if not args.no_solvent:
+        labs.append("FCbulk")
+    if "FREE" in hkldata.df:
+        labs.append("FREE")
+    labs += ll.D_labs + ["S"] # for debugging, for now
+    mtz_out = args.output_prefix+".mtz"
+    hkldata.write_mtz(mtz_out, labs=labs, types={"FOM": "W", "FP":"F", "SIGFP":"Q", "I":"J", "SIGI":"Q", "F_est": "F", "F_exp": "F"})
+
+# main()
+
+if __name__ == "__main__":
+    import sys
+    args = parse_args(sys.argv[1:])
+    main(args)

servalcat/refine/spa.py

@@ -0,0 +1,144 @@
+"""
+Author: "Keitaro Yamashita, Garib N. Murshudov"
+MRC Laboratory of Molecular Biology
+    
+This software is released under the
+Mozilla Public License, version 2.0; see LICENSE.
+"""
+from __future__ import absolute_import, division, print_function, generators
+import gemmi
+import numpy
+import json
+import scipy.sparse
+from servalcat.utils import logger
+from servalcat import utils
+from servalcat.spa import fofc
+from servalcat.spa import fsc
+from servalcat import ext
+b_to_u = utils.model.b_to_u
+u_to_b = utils.model.u_to_b
+
+def calc_D_and_S(hkldata, lab_obs): # simplified version of fofc.calc_D_and_S()
+    bdf = hkldata.binned_df
+    bdf["D"] = 0.
+    bdf["S"] = 0.
+    for i_bin, idxes in hkldata.binned():
+        Fo = hkldata.df[lab_obs].to_numpy()[idxes]
+        Fc = hkldata.df.FC.to_numpy()[idxes]
+        bdf.loc[i_bin, "D"] = numpy.nansum(numpy.real(Fo * numpy.conj(Fc))) / numpy.sum(numpy.abs(Fc)**2)
+        bdf.loc[i_bin, "S"] = numpy.nanmean(numpy.abs(Fo - bdf.D[i_bin] * Fc)**2)
+# calc_D_and_S()
+
+class LL_SPA:
+    def __init__(self, hkldata, st, monlib, lab_obs, source="electron", mott_bethe=True):
+        assert source in ("electron", "xray")
+        self.source = source
+        self.mott_bethe = False if source != "electron" else mott_bethe
+        self.hkldata = hkldata
+        self.lab_obs = lab_obs
+        self.st = st
+        self.monlib = monlib
+        self.d_min_max = hkldata.d_min_max()
+        self.ll = None
+        self.b_aniso = None
+
+    def refine_id(self):
+        if self.source == "electron":
+            return "ELECTRON MICROSCOPY"
+        return "NON-EM SPA" # does not happen, I guess
+
+    def update_ml_params(self):
+        # FIXME make sure D > 0
+        calc_D_and_S(self.hkldata, self.lab_obs)
+
+    def update_fc(self):
+        if self.st.ncs:
+            st = self.st.clone()
+            st.expand_ncs(gemmi.HowToNameCopiedChain.Short, merge_dist=0)
+        else:
+            st = self.st
+
+        self.hkldata.df["FC"] = utils.model.calc_fc_fft(st, self.d_min_max[0] - 1e-6,
+                                                        monlib=self.monlib,
+                                                        source=self.source,
+                                                        mott_bethe=self.mott_bethe,
+                                                        miller_array=self.hkldata.miller_array())
+
+    def prepare_target(self):
+        pass
+
+    def overall_scale(self, min_b=0.5):
+        k, b = self.hkldata.scale_k_and_b(lab_ref=self.lab_obs, lab_scaled="FC")
+        min_b_iso = self.st[0].calculate_b_aniso_range()[0] # actually min of aniso too
+        tmp = min_b_iso + b
+        if tmp < min_b: # perhaps better only adjust b_iso that went too small, but we need to recalculate Fc
+            logger.writeln("Adjusting overall B to avoid too small value")
+            b += min_b - tmp
+        logger.writeln("Applying overall B to model: {:.2f}".format(b))
+        utils.model.shift_b(self.st[0], b)
+        # adjust Fc
+        k_iso = self.hkldata.debye_waller_factors(b_iso=b)
+        self.hkldata.df["FC"] *= k_iso
+        # adjust Fo
+        self.hkldata.df[self.lab_obs] /= k
+    # overall_scale()
+
+    def calc_target(self): # -LL target for SPA
+        ret = 0
+        for i_bin, idxes in self.hkldata.binned():
+            Fo = self.hkldata.df[self.lab_obs].to_numpy()[idxes]
+            DFc = self.hkldata.df.FC.to_numpy()[idxes] * self.hkldata.binned_df.D[i_bin]
+            S = self.hkldata.binned_df.S[i_bin]
+            ret += numpy.nansum(numpy.abs(Fo - DFc)**2) / S + numpy.log(S) * len(idxes)
+        return ret * 2 # friedel mates
+    # calc_target()
+
+    def calc_stats(self, bin_stats=False):
+        # ignore bin_stats for now. better stats are calculated after refinement
+        stats = fsc.calc_fsc_all(self.hkldata, labs_fc=["FC"], lab_f=self.lab_obs)
+        fsca = fsc.fsc_average(stats.ncoeffs, stats.fsc_FC_full)
+        logger.writeln("FSCaverage = {:.4f}".format(fsca))
+        ret = {"summary": {"FSCaverage": fsca, "-LL": self.calc_target()}}
+        # XXX in fsc object, _full is misleading - it's not full in cross validation mode
+        if "D" in self.hkldata.binned_df and "S" in self.hkldata.binned_df:
+            stats[["D", "S"]] = self.hkldata.binned_df[["D", "S"]]
+        ret["bin_stats"] = stats
+        return ret
+
+    def calc_grad(self, atom_pos, refine_xyz, adp_mode, refine_occ, refine_h, specs):
+        dll_dab = numpy.empty_like(self.hkldata.df[self.lab_obs])
+        d2ll_dab2 = numpy.zeros(len(self.hkldata.df.index))
+        blur = utils.model.determine_blur_for_dencalc(self.st, self.d_min_max[0] / 3) # TODO need more work
+        logger.writeln("blur for deriv= {:.2f}".format(blur))
+        for i_bin, idxes in self.hkldata.binned():
+            D = self.hkldata.binned_df.D[i_bin]
+            S = self.hkldata.binned_df.S[i_bin]
+            Fc = self.hkldata.df.FC.to_numpy()[idxes]
+            Fo = self.hkldata.df[self.lab_obs].to_numpy()[idxes]
+            dll_dab[idxes] = -2 * D / S * (Fo - D * Fc)#.conj()
+            d2ll_dab2[idxes] = 2 * D**2 / S
+
+        if self.mott_bethe:
+            dll_dab *= self.hkldata.d_spacings()**2 * gemmi.mott_bethe_const()
+            d2ll_dab2 *= gemmi.mott_bethe_const()**2
+
+        # we need V for Hessian and V**2/n for gradient.
+        d2ll_dab2 *= self.hkldata.cell.volume
+        dll_dab_den = self.hkldata.fft_map(data=dll_dab * self.hkldata.debye_waller_factors(b_iso=-blur))
+        dll_dab_den.array[:] *= self.hkldata.cell.volume**2 / dll_dab_den.point_count
+        self.ll = ext.LL(self.st, atom_pos, self.mott_bethe, refine_xyz, adp_mode, refine_occ, refine_h)
+        self.ll.set_ncs([x.tr for x in self.st.ncs if not x.given])
+        self.ll.calc_grad_it92(dll_dab_den, blur)
+
+        # second derivative
+        d2dfw_table = ext.TableS3(*self.hkldata.d_min_max())
+        d2dfw_table.make_table(1./self.hkldata.d_spacings(), d2ll_dab2)
+        self.ll.make_fisher_table_diag_fast_it92(d2dfw_table)
+        self.ll.fisher_diag_from_table_it92()
+        #json.dump(dict(b=ll.table_bs, pp1=ll.pp1, bb=ll.bb),
+        #          open("ll_fisher.json", "w"), indent=True)
+        #a, (b,c) = ll.fisher_for_coo()
+        #json.dump(([float(x) for x in a], ([int(x) for x in b], [int(x) for x in c])), open("fisher.json", "w"))
+        #logger.writeln("disabling spec_correction in spa target")
+        if specs is not None:
+            self.ll.spec_correction(specs, use_rr=False)