Difference between revisions of "Tutorial for sorting data stored as numpy to on-resonance R1rho analysis"

From relax wiki
Jump to navigation Jump to search
Line 331: Line 331:
 
]
 
]
 
# De-select for analysis those spins who have bad data
 
# De-select for analysis those spins who have bad data
deselet.all()
+
deselect.all()
 
for sel_spin in sel_ids:
 
for sel_spin in sel_ids:
 
     print("Selecting spin %s"%desel_spin)
 
     print("Selecting spin %s"%desel_spin)

Revision as of 23:58, 14 November 2015

Data background

This is data recorded at 600 and 950 MHz.
This should follow relax-4.0.0.Darwin.dmg installation on a mac, only with GUI.

For each spectrometer frequency, the data is saved in np.arrays

  1. one for the residue number,
  2. one for the rates,
  3. one for the errorbars,
  4. one for the RF field strength.

They can be retrieved also with scipy's loadmat command.

The experiments are on-resonance R1rho, and the rates are already corrected for the (small) offset effect, using the experimentally determined R1.

Specifically, the numpy shapes of the data is:

  1. For 600 MHz
    1. residues (1, 60)
    2. rates (60, 10)
    3. errorbars_rate (60, 10)
    4. RFfields (1, 10)
  1. For 950 Mhz
    1. residues (1, 61)
    2. rates (61, 19)
    3. errorbars_rate (61, 19)
    4. RFfields (1, 19)


An example of the data at the 2 fields is:

Create data files for relax

First prepare data, by running in python.

python 1_prepare_data.py

1_prepare_data.py

File: 1_prepare_data.py

import os
import scipy as sc
import scipy.io
import numpy as np

# Set path
cwd = os.getcwd()

fields = [600, 950]
file_names = ['residues', 'rates', 'errorbars_rate', 'RFfields']

# Store data in dictionary
all_data = {}
all_data['fields'] = fields
all_data['file_names'] = file_names

# Make list of residues and make unique
all_res = []

# Loop over the experiments, collect all data
for field in fields:
    print "\n", field

    # Make a dic inside
    all_data['%s'%field] = {}

    # Construct the path to the data
    path = cwd + os.sep + "Archive" + os.sep + "exp_%s"%field + os.sep + "matrices" + os.sep
    all_data['%s'%field]['path'] = path

    # Collect all filename paths
    field_file_name_paths = []
    for file_name in file_names:
        # Create path name
        file_name_path = path + "%s.mat"%file_name
        field_file_name_paths.append(file_name_path)

        # Load the data
        file_name_path_data = sc.io.loadmat(file_name_path)
        # Extract as numpy
        file_name_path_data_np = file_name_path_data[file_name]
        # And store
        all_data['%s'%field]['%s'%file_name] = file_name_path_data
        all_data['%s'%field]['np_%s'%file_name] = file_name_path_data_np

        print file_name, file_name_path_data_np.shape

        # Collect residues
        if file_name == "residues":
            all_res += list(file_name_path_data_np.flatten())

    # Store
    all_data['%s'%field]['field_file_name_paths'] = field_file_name_paths


# Make list of residues and make unique
all_res_uniq = sorted(list(set(all_res)))
all_data['all_res_uniq'] = all_res_uniq

# Write a sequence file for relax
f = open("residues.txt", "w")
f.write("# Residue_i\n")
for res in all_res_uniq:
    f.write("%s\n"%res)
f.close()

f_exp = open("exp_settings.txt", "w")
f_exp.write("# sfrq_MHz RFfield_kHz file_name\n")

# Then write the files for the rates
for field in all_data['fields']:
    resis = all_data['%s'%field]['np_residues'][0]
    rates = all_data['%s'%field]['np_rates']
    errorbars_rate = all_data['%s'%field]['np_errorbars_rate']
    RFfields = all_data['%s'%field]['np_RFfields'][0]

    print "\nfield: %3.3f"%field
    for i, RF_field_strength_kHz in enumerate(RFfields):
        #print "RF_field_strength_kHz: %3.3f"%RF_field_strength_kHz
        # Generate file name
        f_name = "sfrq_%i_MHz_RFfield_%1.3f_kHz.in"%(field, RF_field_strength_kHz)
        cur_file = open(f_name, "w")
        cur_file.write("# resi rate        rate_err\n")

        exp_string = "%11.7f %11.7f %s\n"%(field, RF_field_strength_kHz, f_name)
        print exp_string,
        f_exp.write(exp_string)

        for j, resi in enumerate(resis):
            rate = rates[j, i]
            error = errorbars_rate[j, i]
            string = "%4d %11.7f %11.7f\n"%(resi, rate, error)
            cur_file.write(string)

        cur_file.close()

f_exp.close()

Run analysis in relax GUI

  • Start relax
  • Then click: "New analysis (Cmd+n)
  • Then click icon for "Relaxation dispersion" -> Next
  • Just accept name for the pipe
  • Open the interpreter: Cmd+p
  • Paste in
import os; os.chdir(os.getenv('HOME') + os.sep + 'Desktop' + os.sep + 'temp'); pwd()
  • Then do
script(file='2_load_data.py')
  • You can scroll through earlier commands with: Cmd+ Arrow up
  • When this is done, quit relax
  • Then to convert all xmgrace files to png
cd grace
python grace2images.py

This should give the images of the data

  • Afterwards you can
  • start relax
  • Open the interpreter: Cmd+p
  • Paste in
import os; os.chdir(os.getenv('HOME') + os.sep + 'Desktop' + os.sep + 'temp'); pwd()
  • Close the interpreter
  • File -> Open relax state -> "temp_state.bz2"

Here comes an error message we have to investigate. But this can just be closed?

Essentially one need to de-select spin 51 before one select the models for R1rho and then start the analysis.


The script under it can be run by

relax 2_load_data.py -t log.txt

2_load_data.py

File: 2_load_data.py

# relax import
from pipe_control.mol_res_spin import spin_loop

# Test if running as script or through GUI.
is_script = False
if not hasattr(cdp, "pipe_type"):
    is_script = True
    # We need to create a data pipe, which will tell relax which type of data we are expecting 
    pipe_name = 'relax_disp'
    pipe_bundle = 'relax_disp'
    pipe.create(pipe_name, pipe_bundle)

# Minimum: Just read the sequence data, but this misses a lot of information.
sequence.read(file='residues.txt', res_num_col=1)

# Name the spins
for cur_spin, mol_name, resi, resn, spin_id in spin_loop(full_info=True, return_id=True, skip_desel=True):
    spin.name(name="HN", spin_id=spin_id)
    # Manually force the model to be R2eff, so plotting can be performed later
    cur_spin.model = "R2eff"

# Name the isotope for field strength scaling.
spin.isotope(isotope='15N')


# Open the settings file
set_file = open("exp_settings.txt")
set_file_lines = set_file.readlines()

for line in set_file_lines:
    if "#" in line[0]:
        continue

    # Get data
    field, RF_field_strength_kHz, f_name = line.split()

    # Assign data
    spec_id = f_name
    relax_disp.exp_type(spectrum_id=spec_id, exp_type='R1rho')

    # Set the spectrometer frequency
    spectrometer.frequency(id=spec_id, frq=float(field), units='MHz')

    # Is in kHz, som convert to Hz
    #http://wiki.nmr-relax.com/Relax_disp.spin_lock_offset%2Bfield
    #http://www.nmr-relax.com/manual/relax_disp_spin_lock_field.html
    disp_frq = float(RF_field_strength_kHz)*1000

    # Set The spin-lock field strength, nu1, in Hz
    relax_disp.spin_lock_field(spectrum_id=spec_id, field=disp_frq)

    # Read the R2eff data
    relax_disp.r2eff_read(id=spec_id, file=f_name, dir=None, disp_frq=disp_frq, res_num_col=1, data_col=2, error_col=3)

    # Is this necessary? The time, in seconds, of the relaxation period.
    #relax_disp.relax_time(spectrum_id=spec_id, time=time_sl)


# Plot data
relax_disp.plot_disp_curves(dir='grace', y_axis='r2_eff', x_axis='disp', num_points=1000, extend_hz=500.0, extend_ppm=500.0, interpolate='disp', force=True)

state.save("temp_state", force=True)


# Do it through script
#if False:
#if True:
if is_script:
    # Deselect spin 51, due to weid data point
    deselect.spin(spin_id=":51@HN", change_all=False)

    import os
    from auto_analyses import relax_disp as aa_relax_disp
    from lib.dispersion.variables import EXP_TYPE_CPMG_DQ, EXP_TYPE_CPMG_MQ, EXP_TYPE_CPMG_PROTON_MQ, EXP_TYPE_CPMG_PROTON_SQ, EXP_TYPE_CPMG_SQ, EXP_TYPE_CPMG_ZQ, EXP_TYPE_LIST, EXP_TYPE_R1RHO, MODEL_B14_FULL, MODEL_CR72, MODEL_CR72_FULL, MODEL_DPL94, MODEL_IT99, MODEL_LIST_ANALYTIC_CPMG, MODEL_LIST_FULL, MODEL_LIST_NUMERIC_CPMG, MODEL_LM63, MODEL_M61, MODEL_M61B, MODEL_MP05, MODEL_NOREX, MODEL_NS_CPMG_2SITE_3D_FULL, MODEL_NS_CPMG_2SITE_EXPANDED, MODEL_NS_CPMG_2SITE_STAR_FULL, MODEL_NS_R1RHO_2SITE, MODEL_NS_R1RHO_3SITE, MODEL_NS_R1RHO_3SITE_LINEAR, MODEL_PARAMS, MODEL_R2EFF, MODEL_TP02, MODEL_TAP03
    # Number of grid search increments.  If set to None, then the grid search will be turned off and the default parameter values will be used instead.
    #GRID_INC = None
    GRID_INC = 21
    # The number of Monte Carlo simulations to be used for error analysis at the end of the analysis.
    MC_NUM = 500
    # Model selection technique.
    MODSEL = 'AIC'
    result_dir_name = os.getcwd()
    # Which models to analyse ?
    #MODELS = [MODEL_R2EFF, MODEL_NOREX, MODEL_DPL94, MODEL_TP02, MODEL_TAP03, MODEL_MP05, MODEL_NS_R1RHO_2SITE]
    #MODELS = [MODEL_R2EFF, MODEL_NOREX]
    # R2EFF shall be skipped here
    #MODELS = [MODEL_NOREX]
    # http://wiki.nmr-relax.com/M61
    MODELS = [MODEL_NOREX, MODEL_M61]

    # Fit, instead of read
    r1_fit = True
    # Go
    aa_relax_disp.Relax_disp(pipe_name=pipe_name, pipe_bundle=pipe_bundle, results_dir=result_dir_name, models=MODELS, grid_inc=GRID_INC, mc_sim_num=MC_NUM, modsel=MODSEL, r1_fit=r1_fit)

Run the same analysis in relax through terminal

The analysis can be performed by

python 1_prepare_data.py
relax 2_load_data.py -t log.txt

Following analysis

First make graphs

cd No_Rex/
./grace2images.py -t EPS,PNG
cat r1.out
cat r1rho_prime.out

cd ..
cd M61
./grace2images.py -t PNG
cat kex.out
cat phi_ex.out 

cat kex.out | grep "None          12"
cat phi_ex.out | grep "None          12"

For model M61, spin 12 is:

Further analysis steps

  • Inspect which residues should not be included in analysis
sel_ids = [
":4@HN",
":5@HN",
":6@HN",
":7@HN",
":10@HN",
":11@HN",
]
# De-select for analysis those spins who have bad data
deselect.all()
for sel_spin in sel_ids:
    print("Selecting spin %s"%desel_spin)
    select.spin(spin_id=desel_spin, change_all=False)
  • Inspect which residues should be analysed together for a clustered/global fit.
cluster_ids = [
":13@N",
":15@N",
":16@N",
":25@N"]

# Cluster spins
for curspin in cluster_ids:
    print("Adding spin %s to cluster"%curspin)
    relax_disp.cluster('model_cluster', curspin)
  • Run analysis again

See also