data/scripts/results_analysis.py

'''
Utility functions for analysis after running pyHXExpress
Includes some functionality that requires pyHDX 
      (my github fork has modifications to allow for replicate data)

'''

import os
import importlib
import sys

# hxex_path = os.path.join('')
# sys.path.append(hxex_path)

import hxex_updating as hxex
import numpy as np, pandas as pd
import config as config
from datetime import datetime

pd.set_option('display.max_columns',None)

now = datetime.now()
date = now.strftime("%d%b%Y")

def hxex_reload():
    importlib.reload(hxex)
    importlib.reload(config)
    hxex.config = config

hxex_reload()

###     PyHDX documentation, Examples
###     https://pyhdx.readthedocs.io/en/stable/examples.html
###
#!pip install pyhdx

import proplot as pplt
from scipy.optimize import lsq_linear
from pathlib import Path
import yaml
from Bio import SeqIO
from collections import defaultdict

from pyhdx.models import Coverage
from pyhdx.plot import peptide_coverage
from pyhdx.batch_processing import StateParser
from pyhdx.fitting import fit_d_uptake
from pyhdx.config import cfg
from pyhdx import read_dynamx, HDXMeasurement
from pyhdx.fitting import fit_rates_half_time_interpolate, fit_rates_weighted_average, fit_gibbs_global
from pyhdx.process import filter_peptides, apply_control, correct_d_uptake
from dask.distributed import Client
from pyhdx.plot import dG_scatter_figure
import pyhdx.plot

from collections import defaultdict

# import warnings
# warnings.simplefilter (action='ignore', )#category=FutureWarning)
# warnings.filterwarnings (action='ignore',)# category=RuntimeWarning,)

# pd.set_option('display.max_columns',None)



def combine_batch_data(dirs):
    # Combine the batch outputs into single dataframe files for metadf, datafits, and fitparams
    metadf_comb = pd.DataFrame()
    datafits_comb = pd.DataFrame()
    fitparams_comb = pd.DataFrame()

    metadf_prefix = "metadf_asrun_"
    datafits_prefix = "data_fits_asrun"
    fitparams_prefix = "fitparamsAll_asrun_"

    for dir in dirs:
        csvfiles = [ f for f in os.listdir(dir) if f[-4:]=='.csv'  ]
        mf = [ m for m in csvfiles if m[:len(metadf_prefix)]==metadf_prefix]
        df = [ d for d in csvfiles if d[:len(datafits_prefix)]==datafits_prefix]
        ff = [ f for f in csvfiles if f[:len(fitparams_prefix)]==fitparams_prefix]

        for f in mf:
            md = pd.read_csv(os.path.join(dir,f))#.drop('Index',axis=1)
            md['original_file'] = f
            metadf_comb = pd.concat([metadf_comb,md])#,ignore_index = True)
        for f in df:
            dd = pd.read_csv(os.path.join(dir,f))
            dd['original_file'] = f
            datafits_comb = pd.concat([datafits_comb,dd])
        for f in ff:
            fd = pd.read_csv(os.path.join(dir,f)).drop('Index',axis=1)
            fd['original_file'] = f
            fitparams_comb = pd.concat([fitparams_comb,fd],ignore_index= True)
        

    metadf_comb = metadf_comb.set_index('Index').sort_values('Index')
    datafits_comb = datafits_comb.sort_values('data_id').reset_index(drop=True)
    fitparams_comb = fitparams_comb.sort_values(['data_id','time_idx','charge','rep','ncurves','nboot']).reset_index(drop=True)

    return metadf_comb, datafits_comb, fitparams_comb



def convert_to_uptakedf(datafit,proj=None):
    # convert datafits into a format that hdexa_to_pyhdx can recognize (for residue level avg uptake calcs in pyHDX)
    # there is an assumption in this that the more protected pop1 goes together and less protected pop2 goes together
    # probably mostly okay, but may need to try to account for actual % populations in deciding what goes together 
    # this needs more consideration if ever more than 2 pops

    TD_time = 1e6
    test = datafit.copy()

    test[['start','end']] = test['peptide_range'].str.split('-',expand=True).astype('int')

    try: 
        test['fracDeut_1'] = test['Dabs_1']/test['max_namides']
    except: pass
    try: 
            test['fracDeut_2'] = test['Dabs_2']/test['max_namides']
    except: test['fracDeut_2'] = np.nan
    #cleanup garbage output - accidental carryover of polymodal data when rolling back to previous number of populations, should be fixed now
    test.loc[test['pop_1'] == 1.0, ['centroid_2','Dabs_2','Dabs_std_2','pop_2','pop_std_2','fracDeut_2']] = np.nan

    pops = test.fit_pops.max()
    testpop = {}
    testpop[1] = test.copy()
    testpop[1]['Protein State']=testpop[1].apply(lambda row: row['sample']+'_pop1',axis=1)
    testpop[1]['Theor Uptake #D'] = testpop[1]['Dabs_1']
    testpop[1]['Conf Interval (#D)'] = testpop[1]['Dabs_std_1']
    testpop[1]['pop'] = testpop[1]['pop_1']
    testpop[1]['pop_std'] = testpop[1]['pop_std_1']
    for pop in range(2,pops+1):
        testpop[pop] = testpop[1].copy() 
        testpop[pop]['Protein State']=testpop[pop].apply(lambda row: row['sample']+'_pop'+str(pop),axis=1)
        testpop[pop].loc[testpop[pop]['fit_pops']==pop,['Theor Uptake #D']] = testpop[pop]['Dabs_'+str(pop)]
        testpop[pop].loc[testpop[pop]['fit_pops']==pop,['Conf Interval (#D)']] = testpop[pop]['Dabs_std_'+str(pop)]
        testpop[pop].loc[testpop[pop]['fit_pops']==pop,['pop']] = testpop[pop]['pop_'+str(pop)]
        testpop[pop].loc[testpop[pop]['fit_pops']==pop,['pop_std']] = testpop[pop]['pop_std_'+str(pop)]

    uptakedf = pd.DataFrame()
    for pop in range(1,pops+1):
        uptakedf = pd.concat([uptakedf,testpop[pop]])

    uptakedf['#D'] = uptakedf['Theor Uptake #D'] * uptakedf['UN_TD_corr'] 
    uptakedf['%D'] = uptakedf['Theor Uptake #D'] / uptakedf['max_namides'] * 100.0
    uptakedf['Deut Time (sec)'] = uptakedf['time']
    uptakedf.loc[uptakedf['time']==TD_time,['Deut Time (sec)']] = 'MAX'
    uptakedf['Sequence'] = uptakedf['peptide']
    uptakedf['Peptide Mass'] = uptakedf['centroid']
    uptakedf['maxD'] = uptakedf['max_namides']
    uptakedf = uptakedf.drop(columns=['start','end'])
    uptakedf['Start'] = uptakedf['start_seq']
    uptakedf['End'] = uptakedf['end_seq']
    if not proj:
         proj = test['sample'].unique()[0]
    uptakedf['Protein'] = proj

    #filter some garbage data based on TD-UN
    uptakedf = uptakedf[uptakedf['UN_TD_corr']<1.05]
    
    return uptakedf



def hdexa_to_pyhdx(data,d_percentage=0.85,protein='protein'):
    ### Function to convert data table exported from HDExaminer to the processed DynamX format pyHDX expects
    ### this will leave any extra columns, but will chop out the MAX time points after processing
    
    drop_first=2

    def _time_to_sec(tp,tpunit):
        return  tp * np.power(60.0,'smh'.find(tpunit[0])) 
    if '# Deut' in data.columns:
        data = data.rename(columns={"# Deut":"#D"})
        data['#D'] = data['#D'].fillna(0.0)
        data['#D'] = data['#D'].astype(float)
    if 'Deut %' in data.columns:
        data = data.rename(columns={"Deut %":"%D"})
        data['%D'] = data['%D'].fillna(0.0)
        data['%D'] = data['%D'].astype(float)
    if 'Deut Time' in data.columns:
        data.loc[data['Deut Time'] == 'FD','Deut Time'] = '1e6s'
        data['time unit'] = data['Deut Time'].str[-1]
        data['Deut Time (sec)'] = data['Deut Time'].str[:-1].astype(float)
        data['Deut Time (sec)'] = data.apply(lambda x: _time_to_sec(tp=x['Deut Time (sec)'],tpunit=x['time unit']),axis=1)
        data.loc[data['Deut Time (sec)'] == 1e6,'Deut Time (sec)'] = 'MAX'
    if 'Protein' not in data.columns:
        data['Protein'] = protein


    pyhdx_cols = ['start', 'end' ,'stop' ,'sequence', 'state', 'exposure' ,'uptake' ,'maxuptake',
                'fd_uptake' ,'fd_uptake_sd' ,'nd_uptake' ,'nd_uptake_sd' ,'rfu', 'protein',
                'modification', 'fragment', 'mhp' ,'center' ,'center_sd' ,'uptake_sd' ,'rt',
                'rt_sd' ,'rfu_sd' ,'_sequence' ,'_start' ,'_stop' ,'ex_residues',
                'uptake_corrected']
    data = data.rename(columns={
                "Protein State":"state",
                "Protein":"protein",
                "Start":"start",
                "End":"end",
                "Sequence":"_sequence",
                "Peptide Mass":"mhp",
                "RT (min)":"rt",
                "Deut Time (sec)":"exposure",
                "maxD":"maxuptake",
                "Theor Uptake #D":"uptake_corrected",
                "#D":"uptake",
                "%D":"rfu",
                "Conf Interval (#D)":"rfu_sd",
                "#Rep":"rep",
                "Confidence":"quality",
                "Stddev":"center_sd",
                #"p"
    })

    missing = list(set(pyhdx_cols)-set(data.columns))
    for mcol in missing:
        data[mcol] = np.nan
        if mcol == "rfu_sd": data[mcol] = 0.05 #set 5% error as dummy value

    data['rfu']=data['rfu']/100.
    data.loc[data['exposure']=="0",'rfu_sd']=0.0
    data['stop']=data['end']+1
    data['sequence']=data["_sequence"].copy()
    data['sequence']=[s.replace("P", "p") for s in data["sequence"]]
    # Find the total number of n terminal / c_terminal residues to remove from pyhdx/process.py
    n_term = np.array([len(seq) - len(seq[drop_first:].lstrip("p")) for seq in data["sequence"]])
    c_term = np.array([len(seq) - len(seq.rstrip("p")) for seq in data["sequence"]])
    data["sequence"] = ["x" * nt + s[nt:] for nt, s in zip(n_term, data["sequence"])]
    data["_start"] = data["start"] + n_term
    data["_stop"] = data["stop"] - c_term
    ex_residues = (np.array([len(s) - s.count("x") - s.count("p") for s in data["sequence"]])* d_percentage)
    data["ex_residues"] = ex_residues
    data["uptake_sd"]=data["center_sd"]
    data["nd_uptake"]=0.0
    data["nd_uptake_sd"]=0.0
    data["modification"]=float("nan")
    data["fragment"]=float("nan")
    # upeps = data[data["exposure"]=="0"]["_sequence"].unique()
    # fpeps = data[data["exposure"]=="MAX"]["_sequence"].unique()
    # good_peps = np.array(list(set(upeps) & set(fpeps)))
    #peps = data["_sequence"].unique()
    states = data["state"].unique()
    data["fd_uptake"]="novalue"
    data["fd_uptake_sd"]="novalue"

    for state in states:
        peps = data[data["state"]==state]["_sequence"].unique()
        for pep in peps:
            try: #may have gotten here without a TD measurement 
                fd_up = data[(data["_sequence"]==pep) & (data["exposure"]=="MAX")& (data["state"]==state)]['uptake'].iat[0]
                fd_up_sd = data[(data["_sequence"]==pep) & (data["exposure"]=="MAX")& (data["state"]==state)]['center_sd'].iat[0]
            except: 
                fd_up = data[(data["_sequence"]==pep) & (data["state"]==state)]['maxuptake'].iat[0]
                fd_up_sd = data[(data["_sequence"]==pep) & (data["state"]==state)]['maxuptake'].iat[0]
            data.loc[data["_sequence"]==pep, "fd_uptake"]=fd_up
            data.loc[data["_sequence"]==pep, "fd_uptake_sd"]=fd_up_sd
    data["center"]=data["mhp"]+data["uptake"]
    data["rt_sd"]=0.05 #dummy value

    data['uptake_corrected_orig'] = data['uptake_corrected']
    data['uptake_corrected'] = data["rfu"]*data['maxuptake']

    
    data = data[data["exposure"] != "MAX"]
    data = data[data["fd_uptake"] != 0]
    data = data[~data["uptake"].isna()]
    data["exposure"]=data["exposure"].astype(float)

    new_columns = [col for col in pyhdx_cols if col in data.columns] + [col for col in data.columns if col not in pyhdx_cols]
    return data[new_columns]


def prepare_kwargs(fit_result):
    """Prepare plot kwargs for fit result"""
    d = {
        "y": fit_result.d_uptake.mean(axis=0),
        "fadedata": np.percentile(fit_result.d_uptake, (5, 95), axis=0),
        "shadedata": np.percentile(fit_result.d_uptake, (25, 75), axis=0),
    }
    return d


## Tweaked from pyhdx.plot.single_linear_bar to leave levels in kwargs 
def plot_bar(ax, x, z, cmap, norm, height=1,**kwargs):
    """makes a linear bar plot on supplied axis with values z and corresponding x values x"""

    if isinstance(z, pd.Series):
        z = z.to_numpy()
    elif isinstance(z, pd.DataFrame):
        assert len(z.columns) == 1, "Can only plot dataframes with 1 column"
        z = z.to_numpy().squeeze()

    img = np.expand_dims(z, 0)
    
    collection = ax.pcolormesh(
        pplt.edges(x),
        np.array([0, height]),
        img,
        cmap=cmap,
        vmin=norm.vmin,
        vmax=norm.vmax,
        **kwargs,
    )
    ax.format(yticks=[])

    return collection

from pyhdx.plot import peptide_coverage

def plot_coverage(hdxm,states=None,times=None,savepath=None,peprange=None):
    ## Coverage plots
    use_hdxm = hdxm.copy()
    if states is None:
        states = list(use_hdxm.keys())
    if times is None:
        times = sorted(use_hdxm[states[0]].timepoints)

    fig, axes = pplt.subplots(nrows=len(states),ncols=len(times),  axwidth="200mm", sharey=False, refaspect=2,)

    for j,mutant in enumerate(states):
        for i, use_time in enumerate(times):  
            time_idx = np.where(hdxm[mutant].timepoints == use_time)[0][0]
            filtdf = use_hdxm[mutant][time_idx].data
            if peprange: filtdf = filter_range(filtdf,peprange)
            peptide_coverage(axes[j,i], filtdf,  cbar=True,linewidth=0.1)
            t = axes[j,i].set_title(f'{mutant} Peptides t = {int(use_time)}s')
            l = axes[j,i].set_xlabel('Residue number')
    
    if savepath:
        fig.savefig(savepath,format='pdf',dpi=600)
        
    return


def plot_rfu_residue(hdxm,states=None,times=None,colors=None,savepath=None,legendcols=5):
    # Prolines: 7,12,15,19,38,45,50,51,57,85,124,129,147,154,159,166,172,
    if colors is None:
        colors="#000000 #66C2A5 #56B4E9 #7570B3 #E7298A".split()
    if states is None:
        states = list(hdxm.keys())
    if times is None:
        times = sorted(hdxm[states[0]].timepoints)
    
    z_value ={'50':0.674,'68':1.0,'sd':1.0,'80':1.282,'90':1.645,'95':1.96,'98':2.326,'99':2.576} #confidence intervals multiplier 
  
    nset = 3 #rfu, redundancy, resolution 
    #ncols = 2
    nrows = len(times)*nset
    nfigs = nrows
    array=[]
    for i in range(1,nfigs+1):
        array += [[i,i]]
    hratios = [10,1,1]*len(times)

    fig, axes = pplt.subplots(array,  axheight="30mm",axwidth="180mm",  #refaspect=10,
                                wspace=(0), hspace=([0.7,0.3,3.0]*(len(times)-1)+[0.7,0.3]),
                                sharex=False, sharey=False, hratios=hratios)
    
    vmin = 0
    vmax = 20
    kwargs = dict(levels=pplt.arange(0, vmax*3, 1),) #extendsize=2.5, extendrect=True)
    res_kwargs = dict(levels=pplt.arange(0, vmax, 1),)


    for i, use_time in enumerate(times):

        for j,mutant in enumerate(states):
            time_idx = np.where(hdxm[mutant].timepoints == use_time)[0][0]
            norm = 1#hdxm_hxex[mutant][-1].rfu_residues.values[-1]
            x_res = hdxm[mutant][time_idx].r_number
            center = hdxm[mutant][time_idx].rfu_residues* 1/norm
            #high =  center + z_value['95']*hdxm[mutant][time_idx].rfu_residues_sd
            high50 = center + z_value['50']*hdxm[mutant][time_idx].rfu_residues_sd
            #low = center - z_value['95']*hdxm[mutant][time_idx].rfu_residues_sd
            low50 = center - z_value['50']*hdxm[mutant][time_idx].rfu_residues_sd

            #axes[i*nset].line(hdxm[mutant][time_idx].r_number,hdxm[mutant][time_idx].rfu_residues * 1/norm, label=str(mutant), color=colors[j])
            axes[i*nset].line(x_res,center, shadedata=(low50,high50),label=str(mutant), color=colors[j%len(colors)]) #,fadedata=(low,high)
            axes[i*nset].format(xlabel="",xtickrange=(-1,-1))
            axes[i*nset].format(title=str(use_time)+'s',titleloc= 'lower right',)
        
        redundancy = hdxm[states[0]].coverage.X.sum(axis=0).astype(float)
        resolution = np.repeat(hdxm[states[0]].coverage.block_length, hdxm[states[0]].coverage.block_length)
        resolution = resolution.astype(float)
        resolution[redundancy == 0] = np.nan
        redundancy[redundancy == 0] = np.nan
        red = plot_bar(axes[i*nset+1],hdxm[states[0]].coverage.r_number,redundancy,
                'blues',pplt.Norm("linear", vmin=vmin, vmax=vmax*3),height=1.0,**kwargs)
        axes[i*nset+1].format(xtickrange=(-2,-1))
        res = plot_bar(axes[i*nset+2],hdxm[states[0]].coverage.r_number,resolution, 
                'fire',pplt.Norm("linear", vmin=vmin, vmax=vmax),height=1.0,**res_kwargs)

        axes[i*nset].format(ylabel="D-uptake")
        axes[i*nset+1].set_ylabel("red.",rotation=0,)#loc='center',labelpad=10)
        axes[i*nset+1].yaxis.set_label_coords(-.02,-.05)
        axes[i*nset+1].format(xtickloc='none')
        axes[i*nset+2].set_ylabel("res.",rotation=0,)#loc='center',labelpad=10)
        axes[i*nset+2].yaxis.set_label_coords(-.02,-.05)

    #need to implement https://proplot.readthedocs.io/en/latest/subplots.html?highlight=spacing#Spacing-and-tight-layout
    
    fig.colorbar(red, label="Redundancy (peptides including replicates)",width=0.1,loc='b',length=0.5,col=1,ticks=np.arange(0, vmax*3+1, 5))# **kwargs)
    fig.colorbar(res, label="Resolution (residues)",width=0.1,loc='b',length=0.5,col=2)# **kwargs)
    
    axes[0].legend(loc="t", ncols=legendcols)
    axes.format(ylim=(0,1.1),)#xgrid=True,ygrid=True,ytickrange=(0,1.1))
    axes.format(xlim=(0,180))

    axes[i*nset+2].format(xlabel="Residue number")

    # save_path = os.path.join(project_dir,'B5_phospho_2popFiltShade_RFU_residueavg_plots_'+date+'.pdf')
    if savepath:
        fig.savefig(savepath,format='pdf',dpi=600)

    return #fig

def filter_range(hdxm,peprange):
    df = hdxm.copy()
    peprange = [peprange] if not isinstance(peprange,list) else peprange
    df = df[(df['start'] <= peprange[-1]) & (peprange[0] <= df['end'])]
    return df