analysis.py

import os
import ast
import sys
import math
import heapq
import itertools
import timeit
import itertools
import pickle
from util import *
from collections import OrderedDict
import statium_cpp

def statium(in_res, in_pdb, in_dir, in_ip, out, ip_cutoff_dist, optimize_match, match_cutoffs, backbone, filter_sidechain, counts, verbose):
	
	if verbose: print 'Starting STATUM analysis...' 
	tic = timeit.default_timer()
	sidechain_cpp(in_res, in_pdb, in_dir, in_ip, out, ip_cutoff_dist, optimize_match, match_cutoffs, backbone, filter_sidechain, counts, verbose)
	toc = timeit.default_timer()
	if verbose: print 'Done in ' + str((toc-tic)/60) + ' minutes! Output in: ' + out


def sidechain_cpp(in_res, in_pdb, in_dir, out, ip_dist_cutoff, optimize_match, match_cutoffs, backbone, filter_sidechains, print_counts, verbose):

	#RECEPTOR PEPTIDE
	if verbose: print 'Extracting residue position from ' + in_res + '...'
	res_lines = filelines2list(in_res)
	residues = [int(line.strip())-1 for line in res_lines]
	
	if verbose: print 'Extracting information from ' + in_pdb + '...'
	if filter_sidechains: pdbI = get_pdb_info(in_pdb, filter_sidechains = True)
	else: pdbI = get_pdb_info(in_pdb)
	for res in pdbI:
		if res.string_name == 'GLY':
			res.correct()
			if not backbone:
				res.strip_backbone()
	pdbSize = len(pdbI)	

	if verbose: print 'Computing inter-atomic distances and finding interacting pairs...\n'
	(distance_matrix, use_indices) = dist_matrix_IPs_2terms(pdbI, residues, ip_dist_cutoff)
	if verbose: print use_indices
	num_ips = len(use_indices)

	energies = {key:0 for key in residues}
	for (i, (pos1, pos2)) in enumerate(use_indices):
		aa = pdbI[pos1].char_name
		dists = filter_sc_dists(pdbI[pos1].atom_names, ['CA', 'CB'], distance_matrix[pos1][pos2-pos1-1], 'forward')
		lib_file = os.path.join(in_dir, aa + '.dists')
		result = statium_cpp.query_distance(lib_file, str(dists.values()[1:-1], optimize_match, match_cutoffs[aa]))
		list_result = ast.literal_eval('[' + result + ']')
		for j in range(20): energies[pos2][j] += list_result[j]

	with open(out, 'w') as f:
		f.write('\t\t' + '\t'.join([AAint2char(i) for i in range(20)]))
		for res, energies in energies.iteritems():
			f.write(res + '\t' + '\t'.join(energies))
	
	
def sidechain_python(in_res, in_pdb, in_pdb_dir, in_ip_dir, out, ip_dist_cutoff, match_dist_cutoffs, backbone, filter_sidechains, print_counts, verbose):
	if verbose: print 'Extracting residue position from ' + in_res + '...'
	res_lines = filelines2list(in_res)
	residues = [int(line.strip())-1 for line in res_lines]
	
	if verbose: print 'Extracting information from ' + in_pdb + '...'
	if filter_sidechains: pdbI = get_pdb_info(in_pdb, filter_sidechains = True)
	else: pdbI = get_pdb_info(in_pdb)
	for res in pdbI:
		if res.string_name == 'GLY':
			res.correct()
			if not backbone:
				res.strip_backbone()
	pdbSize = len(pdbI)	

	if verbose: print 'Computing inter-atomic distances and finding interacting pairs...\n'
	(distance_matrix, use_indices) = get_dist_matrix_and_IPs_peptide(pdbI, residues, ip_dist_cutoff)
	if verbose: print use_indices
	num_ips = len(use_indices)
	
	if verbose: print 'Storing distance information for each interacting pair...'
	distances = dict() 
	for (p1,p2) in use_indices:
			if pdbI[p2].stubIntact:
					p2_base_chain = ['CA','CB']
					distances[(p1,p2)] = filter_sc_dists(pdbI[p1].atom_names, p2_base_chain, distance_matrix[p1][p2-p1-1], 'forward') 
			else:
					print 'Position %d does not have a valid stub' % p2
					sys.exit(1)

	#stores total number of each residue identity across IP
	totals = [0 for i in xrange(20)] 
	#stores total number of 'matching sidechain'
	counts = [[0 for j in xrange(20)] for i in xrange(num_ips)]  
	
	lib_pdb_paths = [os.path.join(in_pdb_dir, pdb) for pdb in os.listdir(in_pdb_dir)]
	for i, lib_pdb_path in enumerate(lib_pdb_paths):	
		if verbose: print 'Processing ' + lib_pdb_path + ' (' + str(i) + ' of ' + str(len(lib_pdb_paths)) + ')'
		lib_pdb = get_pdb_info(lib_pdb_path)
		ip_path = os.path.join(in_ip_dir, os.path.split(lib_pdb_path)[1].split('.')[0] + '.ip')
		lib_ips = [(int(pair[0]),int(pair[1])) for pair in filelines2deeplist(ip_path) if pair != []]
		lib_distance_matrix = get_lib_dist_matrix(lib_pdb, lib_ips)
		if not lib_distance_matrix:
				print lib_pdb_path
				continue

		for (lib_pos1, lib_pos2) in lib_ips:
			if abs(lib_pos2 - lib_pos1) <= 4: continue
			
			(lib_AA1, lib_AA2) = (lib_pdb[lib_pos1].int_name, lib_pdb[lib_pos2].int_name)
			if lib_AA1 < 0 or lib_AA2 < 0 or lib_AA1 > 19 or lib_AA2 > 19: continue
			
			totals[lib_AA1] += 1
			totals[lib_AA2] += 1
			
			for (j, (pos1, pos2)) in enumerate(use_indices):
				AA1 = pdbI[pos1].int_name 
				if lib_pdb[lib_pos2].stubIntact:
					if lib_AA1==AA1:
						p2_base_chain = ['CA', 'CB'] 
						lib_dist = filter_sc_dists(lib_pdb[lib_pos1].atom_names, p2_base_chain, lib_distance_matrix[lib_pos1][lib_pos2-lib_pos1-1], True)
						if matching_sidechain_pair(distances[(pos1,pos2)], lib_dist, match_dist_cutoffs[pdbI[pos1].char_name]):
							counts[j][lib_AA2] += 1
						
				if lib_pdb[lib_pos1].stubIntact:
					if lib_AA2==AA1:
						p1_base_chain = ['CA', 'CB'] 
						lib_dist = filter_sc_dists(p1_base_chain, lib_pdb[lib_pos2].atom_names, lib_distance_matrix[lib_pos1][lib_pos2-lib_pos1-1], False) 
						if matching_sidechain_pair(distances[(pos1,pos2)], lib_dist, match_dist_cutoffs[pdbI[pos1].char_name]):
							counts[j][lib_AA1] += 1
	
	if(verbose): print('Finished processing library .pdb files.')

	if print_counts:
		if verbose: print 'Writing raw library counts to files...'

		counts_dir = out + '_counts'
		if not os.path.exists(counts_dir):
			os.mkdir(counts_dir)

		total_counts_file = open(os.path.join(counts_dir, 'lib_ip_residue_totals.txt'), 'w')
		for i in xrange(20): total_counts_file.write(AAint2char(i) + '\t' + str(totals[i]) + '\n')
		total_counts_file.close()

		for (i, pair) in enumerate(use_indices):
			counts_file = open(os.path.join(counts_dir, str(pair[0] + 1) + '_' + str(pair[1] + 1) + '_counts.txt'), 'w')
			for j in xrange(20): counts_file.write(AAint2char(j) + '\t' + str(counts[i][j]) + '\n')
			counts_file.close()
	
	if verbose: print("Computing probabilities from counts...")
	probs = determine_probs(use_indices, totals, counts)

	if len(use_indices) != len(probs):
		print 'Number of IPs is not consistent'
		print use_indices
		print probs
		sys.exit(1)

	write_output(use_indices, probs, out)	
	if(verbose): print("Finished calculating probabilities. Written to: " + out + '_probs')


def get_dist_matrix_and_IPs_peptide(pdb, residues, cutoff):
	N = len(pdb)
	distance_matrix = [[None]*(N-i-1) for i in xrange(N)]
	ips = set()
	first = True
	print '\tOut of %d residues finished:' % N
	for i in xrange(N):
		for j in xrange(i+1, N):
			if (i in residues) ^ (j in residues):
				result = pdb[i].fastFilteredDistancesTo(pdb[j], cutoff)
				distance_matrix[i][j-i-1] = result
				if result is not None:
					if i in residues:
						ips.add((j,i))
					else:
						ips.add((i,j))

	return (distance_matrix, ips)

def dist_matrix_IPs_2terms(pdb, residues, cutoff):
	N = len(pdb)
	distance_matrix = [[None]*(N-i-1) for i in xrange(N)]
	ips = set()
	first = True
	print '\tOut of %d residues finished:' % N
	for i in xrange(N):
		for j in xrange(i+1, N):
                        if j-i <=4: continue
                        result = pdb[i].fastFilteredDistancesTo(pdb[j], cutoff)
                        distance_matrix[i][j-i-1] = result
                        if result is not None:
                                if i in residues:
                                        ips.add((j,i))
                                else:
                                        ips.add((i,j))

	return (distance_matrix, ips)

def get_lib_dist_matrix(pdb, ips):
	N = len(pdb)
	matrix = [[None]*(N-i-1) for i in xrange(N)]

	for (i,j) in ips:
		try:
			matrix[i][j-i-1] = pdb[i].fastDistancesTo(pdb[j])
		except IndexError:
			print 'Invalid indices'
			print i,j
			print len(matrix), len(matrix[0])
			print len(pdb)
			return None
	return matrix

def write_output(use_indices, probs, out):
	AAs = [AAint2char(i) for i in xrange(20)]
	with open(out, 'w') as f:
		s = 'IPs\t' + '\t'.join(AAs)
		f.write(s + '\n')
		for l, (i,j) in enumerate(use_indices):
			s = str(i) + '-' + str(j)
			s2 = '\t'.join([str(probs[l][aa]) for aa in AAs])
			f.write(s + '\t' + s2 + '\n')

def read_output(in_path):
	lines = filelines2deeplist(in_path, skipComments=True, skipEmptyLines=True)
	AAs = [AAint2char(i) for i in xrange(20)]
	probs = OrderedDict()
	for ip_line in lines[1:]:
		ip = tuple(ip_line[0].split('-'))
		use_indices.append(ip)
		aa_probs = dict()
		for i, prob in enumerate(ip_line[1:]):
			aa_probs[AAs[i]] = prob
		probs[ip] = aa_probs

	return probs

def determine_probs(use_indices, totals, counts):
	'''Converts library counts to energies
	Takes in interacting pairs, total IP counts (list of length 20), and AA count frequences for each peptide IP
	Energies are output as list of dicts (key:value as AA character:energy value), one dict per IP'''
	#total number of residues across all library interacting pairs
	lib_sum = float(sum(totals))

	#frequency of each residue in total library
	lib_total_probs = [x/lib_sum for x in totals]
	out = list()
	
	for (i, pair) in enumerate(use_indices):
		total = float(sum(counts[i]))
		if total > 99:
			probs = dict()
			#probability of each residue occuring at *that* IP (i.e. / by total res's at the IP)
			AA_probs = [(x/total if x != 0 else 1/total) for x in counts[i]]
			for j in xrange(20):
				e = -1.0 * math.log(AA_probs[j] / lib_total_probs[j])
				probs[AAint2char(j)] = e
			out.append(probs)
	return out

#Assumes fast & distances stored as dict (atom1_name, atom2_name): distance val
def matching_sidechain_pair(dists1, dists2, cutoff):
  
	sd = 0.0
	count = 0.0

	for pair1, dist1 in dists1.iteritems():
		for pair2, dist2 in dists2.iteritems():
			if pair1 == pair2:
				sd += ((dist1 - dist2) ** 2)
				count += 1.0

	if math.sqrt(sd / count) < cutoff: return True
	else: return False

	

#returns a more sparse distance matrix, filled on at IP positions
#	uses the library pdb_info
def get_distance_matrix_ip(pdb, ips):
	
	N = len(pdb)
	distance_matrix = [[None]*N for j in xrange(N)]
	
	for (i,j) in ips:
		try:
			distance_matrix[i][j] = pdb[i].distancesTo(pdb[j])
		except:
			print len(pdb)
			print pdb
			print i,j,N
			
	return distance_matrix
	

#If fast, assumes pos1_list is atom_names (not atom objects), and pair_dists is indexed the same way
def filter_sc_dists(pos1_list, pos2_list, pair_dists, forward, output_dict=True, fast=True):
  
	if output_dict:
			out = dict()	
			if forward:
				for atomi in pos1_list:
					for atomj in pos2_list:
							out[(atomi, atomj)] = pair_dists[(atomi, atomj)]
			else:
				for atomj in pos2_list:
					for atomi in pos1_list:
							out[(atomj, atomi)] = pair_dists[(atomi, atomj)]

			return out

	else:
			pairs = list()
			dists = list()

			if forward:
				for atomi in pos1_list:
					for atomj in pos2_list:
							idx = distance_matrix[0].index((atomi, atomj))
							pairs.append((atomi, atomj))
							dists.append(distance_matrix[1][idx])
			else:
				for atomj in pos2_list:
					for atomi in pos1_list:
							idx = distance_matrix[0].index((atomi, atomj))
							pairs.append((atomj, atomi))
							dists.append(distance_matrix[1][idx])

			return [pairs, dists]



def generate_random_distribution (in_res, in_probs_dir, num_seqs=1000):
	
	sequence_length = len(filelines2list(in_res))

	print 'Calculating distribution of energies for %d random sequences...', num_seqs
	energies = list()
	for i in xrange(num_seqs):
		if i % 100 is 0:
			print i,
			sys.stdout.flush()
		energies.append(calc_seq_energy(in_res, in_probs_dir, generate_random_seq(sequence_length)))
		
	energies.sort()
	avg = mean(energies)
	sd = std(energies)
		
	return (sequence_length, energies, avg, sd)


def calc_seq_energy (in_res_path, in_probs, seq):

	#Handle two cases: e.g. AAAX,LLL and LXAAM
	seq = ''.join(seq.split(','))
	
	#loading in probability into all_probs
	all_probs = read_output(in_probs)
	
	energy = 0
	lines = filelines2list(in_res_path)
	residues = [int(line.strip()) for line in lines]

	for i, residue in enumerate(residues):
		filtered = {ip: probs for ip, probs in all_probs.items() if ip[1] == residue}
		AA = seq[i-1]
		if AA == 'X' or AA == 'G' or filtered == {}: continue
		for ip, probs in filtered.items():
			energy += probs[AAchar2int(AA)]
			
	return energy
	
#Note: need res file, because not all residues 
def calc_top_seqs(in_res_path, in_probs, num_sequences):
	 
	#read back from .res file where ligand residues start
	lines = filelines2list(in_res_path)
	residues = [int(line.strip()) for line in lines]
	
	#loading in probability into all_probs
	prob_files = os.listdir(probs_dir)
	all_probs = read_output(in_probs)
	
	#the following now fills ordered_probs
	#[[sorted list of AA probs for a residue position: (0.5, 'A'), (0.3, 'C'),...], [like before for residue pos 2], etc...]
	ordered_probs = [] 
	AAs = ['A', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'V', 'W', 'Y']
	for residue in residues:
		probs = [prob for ip, prob in all_probs.items() if residue in ip]
		probs_sum = map(sum, zip(*probs))
		sorted_probs = sorted(zip(probs_sum, AAs), key=lambda pair: pair[0])
		ordered_probs.append(sorted_probs)
	
	#enumerate all the balls in urns possibilities
	#note that to maintain the max-heap property in Python's heapq implementation (which only does min-heap), we multiple by -1 before adding to heap
	num_urns = len(residues)
	heap = []
	
	seq = ''
	energy = 0
	for i, c in enumerate(residues):
		if ordered_probs[i]:
			aa = ordered_probs[i][0][1]
			seq += aa
			energy += (0.0 if aa == 'G' else ordered_probs[i][0][0])
		else:
			seq += 'X'
	heapq.heappush(heap, (energy, seq))
	
	max_num_balls = 0
	total = 0
	while total < num_sequences:
		max_num_balls += 1
		total += nCr((max_num_balls+num_urns-1), (max_num_balls))
	
	for num_balls in xrange(max_num_balls+1)[1:]:
		combo_elements = xrange(num_balls+num_urns-1)
		combos = list(itertools.combinations(combo_elements, (num_urns-1)))
				
		for combo in combos:
			#print combo
			urn_counts = [0]*num_urns
			for i, position in enumerate(combo):
				if(i == 0):
					urn_counts[i] = position
				elif(i == len(combo)-1):
					urn_counts[i] = position - combo[i-1] - 1
					urn_counts[i+1] = len(combo_elements)-1-position
				else:
					urn_counts[i] = position - combo[i-1] - 1
			
				seq = ''
				energy = 0
				for i, c in enumerate(urn_counts):
					if(ordered_probs[i]):
						aa = ordered_probs[i][c][1]
						seq += aa
						energy += (0.0 if aa == 'G' else ordered_probs[i][c][0])
					else:
						seq += 'X'
			
			if(seq in [i[1] for i in heap]):
				continue
				
			if(len(heap) < num_sequences):
				heapq.heappush(heap, (-1*energy, seq))
			elif(energy < heap[0][0]*-1):
				heapq.heappushpop(heap, (-1*energy, seq))
	
	heap = sorted(heap, reverse=True)
	out = [(seq, energy*-1) for energy, seq in heap]
	return out

