evaluate.py 24.2 KB
Newer Older
1
2
3
4
5
6
7
#!/usr/bin/env python3


import sys
import argparse
from termcolor import colored
import networkx as nx
8
sys.setrecursionlimit(10000)
9
10
11


def parse_args():
12
    parser = argparse.ArgumentParser(description='Process a d2 graph (complete graph or path) to evaluate its quality.')
13
    parser.add_argument('filename', type=str,
Yoann Dufresne's avatar
Yoann Dufresne committed
14
                        help='The file to evalute')
Yoann Dufresne's avatar
Yoann Dufresne committed
15
    parser.add_argument('--type', '-t', choices=["d2", "path", "d2-2annotate", "dgraphs"], default="path", required=True,
16
                        help="Define the data type to evaluate. Must be 'd2' or 'path' or 'd2-2annotate' (Rayan's hack).")
Yoann Dufresne's avatar
Yoann Dufresne committed
17
18
    parser.add_argument('--light-print', '-l', action='store_true',
                        help='Print only wrong nodes and paths')
19
    parser.add_argument('--max_gap', '-g', type=int, default=0, help="Allow to jump over max_gap nodes during the increasing path search")
Yoann Dufresne's avatar
Yoann Dufresne committed
20
    parser.add_argument('--barcode_graph', '-b', help="Path to the barcode graph corresponding to the d2_graph to analyse.")
21
    parser.add_argument('--optimization_file', '-o',
Yoann Dufresne's avatar
Yoann Dufresne committed
22
                        help="If the main file is a d2, a file formatted for optimization can be set. This file will be used to compute the coverage of the longest path on the barcode graph.")
23
24
25
26
27

    args = parser.parse_args()
    return args


Yoann Dufresne's avatar
Yoann Dufresne committed
28
29
30
31
32
33
34
35
36
37
38
39
40
41
def transform_bg(graph):
    idx = 0
    node_names = {}
    nx.set_node_attributes(graph, 0, 'score')
    nx.set_node_attributes(graph, "", 'barcode_edges')
    for node in graph.nodes():
        graph.nodes[node]['udg'] = f"[{node}][][]"
        node_names[node] = str(idx)
        idx += 1

    graph = nx.relabel_nodes(graph, node_names)

    return graph

42
43


44

Yoann Dufresne's avatar
Yoann Dufresne committed
45
46
# ------------- Path Graph -------------

Yoann Dufresne's avatar
Yoann Dufresne committed
47
48
49
50
def mols_from_node(node_name):
    return [int(idx) for idx in node_name.split(":")[1].split(".")[0].split("_")]


51
52
53
54
def parse_udg_qualities(graph):
    """ Compute the quality for the best udgs present in the graph.
        All the node names must be under the format :
        {idx}:{mol1_id}_{mol2_id}_...{molx_id}.other_things_here
55
56
        :param graph: The networkx graph representing the deconvolved graph
        :return: A tuple containing two dictionaries. The first one with theoretical frequencies of each node, the second one with observed frequencies.
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
    """
    dg_per_node = {}

    for node, data in graph.nodes(data=True):
        str_udg = data["udg"]
        central, h1, h2 = str_to_udg_lists(str_udg)

        if central not in dg_per_node:
            dg_per_node[central] = []
        dg_per_node[central].append(data["udg"])

    for node in dg_per_node:
        print(node, dg_per_node[node])
        print(len(dg_per_node))

    return dg_per_node


Yoann Dufresne's avatar
Yoann Dufresne committed
75
def parse_path_graph_frequencies(graph, barcode_graph):
76
77
78
79
    """ Compute appearance frequencies from node names.
        All the node names must be under the format :
        {idx}:{mol1_id}_{mol2_id}_...{molx_id}.other_things_here
        :param graph: The networkx graph representing the deconvolved graph
80
        :param barcode_graph: The barcode graph
81
82
        :return: A tuple containing two dictionaries. The first one with theoretical frequencies of each node, the second one with observed frequencies.
    """
Yoann Dufresne's avatar
Yoann Dufresne committed
83
    # Compute origin nodes formatted as `{idx}:{mol1_id}_{mol2_id}_...`
Yoann Dufresne's avatar
Yoann Dufresne committed
84
    observed_frequencies = {}
Yoann Dufresne's avatar
Yoann Dufresne committed
85
    real_frequencies = {}
86
    origin_node_names = []
Yoann Dufresne's avatar
Yoann Dufresne committed
87
88
    node_per_barcode = {}

Yoann Dufresne's avatar
Yoann Dufresne committed
89
    for node, data in graph.nodes(data=True):
90
91
        parsed = parse_dg_name(graph, node)
        origin_name = parsed[0][1]
Yoann Dufresne's avatar
Yoann Dufresne committed
92
93

        if origin_name not in node_per_barcode:
Yoann Dufresne's avatar
Yoann Dufresne committed
94
95
            node_per_barcode[origin_name] = []
        node_per_barcode[origin_name].append(node)
96
97

        # Count frequency
Yoann Dufresne's avatar
Yoann Dufresne committed
98
        if origin_name not in observed_frequencies:
Yoann Dufresne's avatar
Yoann Dufresne committed
99
            observed_frequencies[origin_name] = 0
100
            origin_node_names.append(origin_name)
Yoann Dufresne's avatar
Yoann Dufresne committed
101
        observed_frequencies[origin_name] += 1
102

Yoann Dufresne's avatar
Yoann Dufresne committed
103
104
    # Theoretical frequencies
    real_frequencies = {node_id: len(node_id.split(":")[1].split("_")) for node_id in barcode_graph.nodes()}
105

Yoann Dufresne's avatar
Yoann Dufresne committed
106
    return real_frequencies, observed_frequencies, node_per_barcode
107
108


Yoann Dufresne's avatar
Yoann Dufresne committed
109
def parse_graph_path(graph):
110
111
112
    """ This function aims to look for direct molecule neighbors.
        If a node has more than 2 direct neighbors, it's not rightly split
    """
Yoann Dufresne's avatar
Yoann Dufresne committed
113
114
115
116
117
    neighborhood = {}

    for node in graph.nodes():
        molecules = mols_from_node(node)
        neighbors = list(graph.neighbors(node))
118

Yoann Dufresne's avatar
Yoann Dufresne committed
119
120
121
122
123
124
125
126
        neighborhood[node] = []
        for mol in molecules:
            for nei in neighbors:
                nei_mols = mols_from_node(nei)
                if mol-1 in nei_mols:
                    neighborhood[node].append(nei)
                if mol+1 in nei_mols:
                    neighborhood[node].append(nei)
127

Yoann Dufresne's avatar
Yoann Dufresne committed
128
    return neighborhood
129
130


Yoann Dufresne's avatar
Yoann Dufresne committed
131
132
def print_path_summary(frequencies, light_print=False, file_pointer=sys.stdout):
    if file_pointer is None:
Yoann Dufresne's avatar
Yoann Dufresne committed
133
134
        return

Yoann Dufresne's avatar
Yoann Dufresne committed
135
    print("--- Nodes analysis ---", file=file_pointer)
Yoann Dufresne's avatar
Yoann Dufresne committed
136
137
138
    theoretical_frequencies, observed_frequencies, node_per_barcode = frequencies
    for key in theoretical_frequencies:
        obs, the = observed_frequencies[key] if key in observed_frequencies else 0, theoretical_frequencies[key]
Yoann Dufresne's avatar
Yoann Dufresne committed
139
140
141
142
143
        result = f"{key}: {obs}/{the}"

        if file_pointer == sys.stdout:
            result = colored(result, 'green' if obs==the else 'red')

Yoann Dufresne's avatar
Yoann Dufresne committed
144
        if light_print and obs == the:
Yoann Dufresne's avatar
Yoann Dufresne committed
145
146
147
148
149
            continue

        print(result, file=file_pointer)


150
151
    print("--- Global summary ---", file=file_pointer)

Yoann Dufresne's avatar
Yoann Dufresne committed
152
153
    # --- Frequency usage ---
    # Tags
Yoann Dufresne's avatar
Yoann Dufresne committed
154
155
156
    distinct_theoretical_nodes = len(theoretical_frequencies)
    distinct_observed_nodes = len(observed_frequencies)
    print(f"Distinct barcodes: {distinct_observed_nodes}/{distinct_theoretical_nodes}", file=file_pointer)
Yoann Dufresne's avatar
Yoann Dufresne committed
157
    # molecules
Yoann Dufresne's avatar
Yoann Dufresne committed
158
159
160
    cumulative_theoretical_nodes = sum(theoretical_frequencies.values())
    cumulative_observed_nodes = sum(observed_frequencies.values())
    print(f"Molecules: {cumulative_observed_nodes}/{cumulative_theoretical_nodes}", file=file_pointer)
Yoann Dufresne's avatar
Yoann Dufresne committed
161
162
163
    # Wrong splits
    over_split = 0
    under_split = 0
Yoann Dufresne's avatar
Yoann Dufresne committed
164
165
166
167
168
    for barcode in theoretical_frequencies:
        observed = observed_frequencies[barcode] if barcode in observed_frequencies else 0
        theoretic = theoretical_frequencies[barcode]
        over_split += max(observed-theoretic, 0)
        under_split += max(theoretic-observed, 0)
Yoann Dufresne's avatar
Yoann Dufresne committed
169
    print(f"Under/Over splitting: {under_split} - {over_split}")
170
171


172
173
174
175
def print_dgraphs_summary(frequencies, light_print=False):
    pass


Yoann Dufresne's avatar
Yoann Dufresne committed
176
177
# ------------- D2 Graph -------------

178
179
def str_to_udg_lists(s):
    udg = s.replace("]", "").replace(' [', '[')
180
    return udg.split('[')[1:]
181

Rayan  CHIKHI's avatar
Rayan CHIKHI committed
182
183
184
185
# speeds up networkx access to attributes
cached_udg_attr = None
cached_score_attr = None

186
def parse_dg_name(gr, name):
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
187
188
189
190
    global cached_udg_attr, cached_score_attr 
    if cached_udg_attr is None:
        cached_udg_attr = nx.get_node_attributes(gr, 'udg')
    udg = cached_udg_attr[name]
Rayan Chikhi's avatar
Rayan Chikhi committed
191
192
193
194
    res = str_to_udg_lists(udg)
    if len(res) != 3:
        print("parsing problem:",res)
    central, h1, h2 = res
Yoann Dufresne's avatar
Yoann Dufresne committed
195
    
196
    idx = name
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
197
198
199
    if cached_score_attr is None:
        cached_score_attr = nx.get_node_attributes(gr, 'score')
    score = cached_score_attr[name]
Yoann Dufresne's avatar
Yoann Dufresne committed
200
201
202
203
204
205
206
207

    # Parse hands
    h1 = h1.split(', ')
    h2 = h2.split(', ')

    return (idx, central, score), h1, h2


208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
def path_to_jumps(path):
    chuncks = []
    prev_start = -1000
    current_molecule = -1000

    for mol, node in path:
        # If there is a gap
        if mol > current_molecule + 1:
            chuncks.append((prev_start, current_molecule))
            prev_start = mol

        current_molecule = mol

    # Add the last piece
    chuncks.append((prev_start, current_molecule))

    del chuncks[0]
    return chuncks


Yoann Dufresne's avatar
Yoann Dufresne committed
228
def print_d2_summary(connected_components, longest_path, coverage_vars=(0, 0), light_print=False):
Yoann Dufresne's avatar
Yoann Dufresne committed
229
230
231
    print("--- Global summary ---")
    print(f"Number of connected components: {len(connected_components)}")
    print(f"Total number of nodes: {sum([len(x) for x in connected_components])}")
Yoann Dufresne's avatar
Yoann Dufresne committed
232
233
    no_singleton = [x for x in connected_components if len(x) > 1]
    print(f"There are {len(no_singleton)} connex components with at least 2 nodes")
Yoann Dufresne's avatar
Yoann Dufresne committed
234
235
236
    print(f"The 5 largest components: {[len(x) for x in connected_components][:5]}")

    print("--- Largest component analysis ---")
237
238
239
240
241
242
243
    # Get the list of node idx
    path_dg_idx = [int(x[1].split(" ")[0]) for x in longest_path]
    # print("\n".join(longest_path))
    if not light_print:
        print("Longest path for increasing molecule number:")
        print(path_dg_idx)
    print(f"Size of the longest path: {len(longest_path)}")
Yoann Dufresne's avatar
Yoann Dufresne committed
244
245
246
    if not light_print:
        print("Jumps in central nodes:")
        print(path_to_jumps(longest_path))
Yoann Dufresne's avatar
Yoann Dufresne committed
247
248
249

    print(f"Number of usable coverage variables: {len(coverage_vars[1])}")
    print(f"Coverage: {len(coverage_vars[0])}/{len(coverage_vars[1])}")
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
    if not light_print:
        print(f"Missing coverage variables:\n{coverage_vars[1]-coverage_vars[0]}")

def _get_distant_neighbors(graph, node, dist):
    neighbors = set()

    to_compute = [node]
    for _ in range(dist):
        next_compute = []
        for node in to_compute:
            for neighbor in graph[node]:
                if neighbor not in neighbors:
                    neighbors.add(neighbor)
                    next_compute.append(neighbor)
        to_compute = next_compute
Yoann Dufresne's avatar
Yoann Dufresne committed
265

266
    return neighbors
Yoann Dufresne's avatar
Yoann Dufresne committed
267

268
def compute_next_nodes(d2_component, max_jumps=0):
269
270
271
272
273
    # First parse dg names
    dg_names = {}
    for node in d2_component.nodes():
        dg_names[node] = parse_dg_name(d2_component,node)

Yoann Dufresne's avatar
Yoann Dufresne committed
274
275
276
277
    next_nodes = {}

    for node in d2_component.nodes():
        # Parse the current node name
278
        head, h1, h2 = dg_names[node]
Yoann Dufresne's avatar
Yoann Dufresne committed
279
280
281
282
        next_nodes[node] = {}

        # Get the current molecule idxs
        molecule_idxs = mols_from_node(head[1])
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
283
284
        #print("node",node,"dg name",dg_names[node],"mol idxs",molecule_idxs)

Yoann Dufresne's avatar
Yoann Dufresne committed
285
286
        for mol_idx in molecule_idxs:
            nexts = []
287
288
            # for neighbor in d2_component[node]:
            for neighbor in _get_distant_neighbors(d2_component, node, max_jumps+1):
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
289
                # nei_head: central node of the neighbor of 'node'
290
                nei_head, _, _ = dg_names[neighbor]
Yoann Dufresne's avatar
Yoann Dufresne committed
291
                nei_mols = mols_from_node(nei_head[1])
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
292
                # only consider neighbor molecules that are strictly bigger than the current molecule idx considered (from 'node')
Yoann Dufresne's avatar
Yoann Dufresne committed
293
294
295
296
297
                nei_mols = [x for x in nei_mols if x > mol_idx]
                
                # If there are molecule next
                if len(nei_mols) > 0:
                    next_nei_mol = min(nei_mols)
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
298
                    # append to the neighbors of (node,mol_idx) that 'neighbor' if it contains a molecule that's bigger than mol_idx
Yoann Dufresne's avatar
Yoann Dufresne committed
299
300
301
302
                    nexts.append((next_nei_mol, neighbor))

            nexts.sort(key=lambda x: x[0])
            next_nodes[node][mol_idx] = nexts
Yoann Dufresne's avatar
Yoann Dufresne committed
303
            # print("next nodes of node",node,"mol idx",mol_idx,":",next_nodes)
Yoann Dufresne's avatar
Yoann Dufresne committed
304
305
306
307

    return next_nodes


308
309
def compute_longest_increasing_paths(d2_component, max_gap=0):
    next_nodes = compute_next_nodes(d2_component, max_jumps=max_gap)
Yoann Dufresne's avatar
Yoann Dufresne committed
310
    sys.setrecursionlimit(len(d2_component.nodes)*2)
Yoann Dufresne's avatar
Yoann Dufresne committed
311
312
313
314
315
316

    # Compute the longest path for each node
    longest_paths = {}
    for idx, start_node in enumerate(next_nodes):
        # print(f"{idx}/{len(next_nodes)}")
        for mol_idx in next_nodes[start_node]:
317
            recursive_longest_path(start_node, mol_idx, next_nodes, longest_paths)
Yoann Dufresne's avatar
Yoann Dufresne committed
318

Rayan  CHIKHI's avatar
Rayan CHIKHI committed
319
320
    # Get the longest path size,
    # across all barcode graph nodes and all molecules in these barcodes
Yoann Dufresne's avatar
Yoann Dufresne committed
321
322
323
324
325
326
327
328
329
330
331
332
333
334
    max_len, node_val, mol_idx = 0, None, -1
    for node in longest_paths:
        for mol in longest_paths[node]:
            length, _, _ = longest_paths[node][mol]
            if max_len < length:
                max_len = length
                node_val = node
                mol_idx = mol

    # Backtrack the longest path
    path = backtrack_longest_path(node_val, mol_idx, longest_paths)
    return path


Yoann Dufresne's avatar
Yoann Dufresne committed
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
def compute_shortest_edit_path(path):
    min_mol = float("inf")
    max_mol = 0
    molecules = {}

    path_extremes = []
    node_names = {}

    # Parse molecules
    for node_name, data in path.nodes(data=True):
        udg = data["udg"]
        central_node = udg.split(']')[0][1:]

        node_names[node_name] = central_node
        if len(list(path[node_name])) == 1:
            path_extremes.append(node_name)

        mol_names = central_node.split(":")[1].split('_')
        for mol_name in mol_names:
            mol_idx = int(mol_name)
            if mol_idx < min_mol: min_mol = mol_idx
            if mol_idx >= max_mol: max_mol = mol_idx
            molecules[mol_idx] = central_node

    # create barcode path from molecules
    molecule_order = []
    for idx in range(min_mol, max_mol+1):
        if idx in molecules:
            molecule_order.append(molecules[idx])

    # create barcode path from d2_path
    first = path_extremes[0]
    last = path_extremes[1]
    d2_path_order = [first, list(path[first])[0]]
    while d2_path_order[-1] != last:
        neighbors = list(path[d2_path_order[-1]])

        if neighbors[0] == d2_path_order[-2]:
            d2_path_order.append(neighbors[1])
        else:
            d2_path_order.append(neighbors[0])
    d2_path_order = [node_names[x] for x in d2_path_order]

    edit_path = edit_distance(molecule_order, d2_path_order)
    filtered_edit_path = [x[0] for x in edit_path if x[0] == x[1]]
    reverse_edit_path = edit_distance(molecule_order, d2_path_order[::-1])
Yoann Dufresne's avatar
Yoann Dufresne committed
381
    reverse_filtered_edit_path = [x[0] for x in reverse_edit_path if x[0] == x[1]]
Yoann Dufresne's avatar
Yoann Dufresne committed
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437

    if len(filtered_edit_path) > len(reverse_filtered_edit_path):
        return filtered_edit_path
    else:
        return reverse_filtered_edit_path



def edit_distance(array_vertical, array_horizontal):
    dynamic = [[float("inf") for column in range(len(array_horizontal)+1)] for row in range(len(array_vertical)+1)]
    # Fill init line and column
    for i in range(len(array_horizontal)+1):
        dynamic[0][i] = i
    for i in range(len(array_vertical)+1):
        dynamic[i][0] = i

    # Fill the array
    for row in range(1, len(array_vertical)+1):
        for column in range(1, len(array_horizontal)+1):
            if array_vertical[row-1] == array_horizontal[column-1]:
                dynamic[row][column] = min(dynamic[row-1][column-1], dynamic[row-1][column]+1, dynamic[row][column-1]+1)
            else:
                dynamic[row][column] = min(dynamic[row-1][column-1]+1, dynamic[row-1][column]+1, dynamic[row][column-1]+1)

    # Compute alignment
    row = len(array_vertical)
    column = len(array_horizontal)
    path = [(array_vertical[row-1], array_horizontal[column-1])]

    while row != 0 and column != 0:
        score = dynamic[row][column]
        # Match
        if array_vertical[row-1] == array_horizontal[column-1] and dynamic[row-1][column-1] == dynamic[row][column]:
            row -= 1
            column -= 1
        elif array_vertical[row-1] != array_horizontal[column-1] and dynamic[row-1][column-1] == dynamic[row][column] - 1:
            row -= 1
            column -= 1
        elif dynamic[row-1][column] == dynamic[row][column] - 1:
            row -= 1
        elif dynamic[row][column-1] == dynamic[row][column] - 1:
            column -= 1
        else:
            print("Huge problem in edit distance !", file=sys.stderr)
            return []

        path.append((array_vertical[row-1], array_horizontal[column-1]))

    for finalize_row in range(row-1, -1, -1):
        path.append((array_vertical[finalize_row], array_horizontal[column]))
    for finalize_column in range(column-1, -1, -1):
        path.append((array_vertical[row], array_horizontal[finalize_column]))

    return path[::-1]


Yoann Dufresne's avatar
Yoann Dufresne committed
438
def backtrack_longest_path(node, molecule, longest_paths, path=[]):
439
    if node is None:
Yoann Dufresne's avatar
Yoann Dufresne committed
440
441
        return path

442
    path.append((molecule, node))
Yoann Dufresne's avatar
Yoann Dufresne committed
443
444
445
446
447
448
449
    length, next_node, next_mol = longest_paths[node][molecule]
    return backtrack_longest_path(next_node, next_mol, longest_paths, path)


def recursive_longest_path(current_node, current_molecule, next_nodes, longest_paths):
    # Dynamic programming
    if current_node in longest_paths and current_molecule in longest_paths[current_node]:
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
450
        #print("getting cached result for node",current_node,"mol",current_molecule,longest_paths[current_node][current_molecule])
Yoann Dufresne's avatar
Yoann Dufresne committed
451
452
453
454
        return longest_paths[current_node][current_molecule]

    longest = 0
    longest_next = None
455
    min_mol_idx = float('inf') 
Yoann Dufresne's avatar
Yoann Dufresne committed
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
    # Recursively compute the longest path
    for mol_idx, node in next_nodes[current_node][current_molecule]:
        size, _, _ = recursive_longest_path(node, mol_idx, next_nodes, longest_paths)
        if size + 1 > longest:
            longest = size + 1
            longest_next = node
            min_mol_idx = mol_idx
        # If there is an alternative path with shorter distance
        elif size + 1 == longest and mol_idx < min_mol_idx:
            longest = size + 1
            longest_next = node
            min_mol_idx = mol_idx
    
    # Save the result
    if not current_node in longest_paths:
        longest_paths[current_node] = {}
    longest_paths[current_node][current_molecule] = (longest, longest_next, min_mol_idx)
    return longest_paths[current_node][current_molecule]


476

Yoann Dufresne's avatar
Yoann Dufresne committed
477
478
479
480
def compute_covered_variables(graph, path):
    path_nodes = set()
    for mol, node_name in path:
        path_nodes.add(node_name)
481

Yoann Dufresne's avatar
Yoann Dufresne committed
482
483
484
485
486
487
488
    used_vars = set()
    total_vars = set()
    for node, data in graph.nodes(data=True):
        vars = data["barcode_edges"].split(" ")
        total_vars.update(vars)
        if node in path_nodes:
            used_vars.update(vars)
489

Yoann Dufresne's avatar
Yoann Dufresne committed
490
    return used_vars, total_vars
491

492
# returns True iff there exist x in mol1 such that there exists y in mol2 and |x-y| <= some_value
493
494
495
496
497
498
499
500
501
502
503
def nearby_udg_molecules(mols1, mols2):
    for x in mols1:
        for y in mols2:
            if abs(x-y) <= 5:
                return True
    return False

def verify_graph_edges(d2_component):
    udg_molecules_dict=dict()
    for node in d2_component.nodes():
        # Parse the current node name
504
        head, c1, c2 = parse_dg_name(d2_component,node)
505
506

        # Construct the molecule(s) that this udg really 'reflects'
507
508
509
510
511
        # i.e. the udg has a central node and two cliques
        # that central node is the result of merging of several molecules
        # ideally, only one of those molecules is connected to the molecules of the cliques
        # (there could be more than one though; in that case the udg is 'ambiguous')
        # udg_molecules aims to reflect the molecule(s) underlying this udg
512
513
514
515
        udg_molecules = set()

        # Get the current molecule idxs of central node
        molecule_idxs = mols_from_node(head[1])
516
        # print("mol idxs", molecule_idxs)
517
518

        # Examine molecule idx's of cliques to see which are close to the central node
519
        # rationale: c1/c2 contain nearby molecule id's
520
521
522
523
524
        for mol_idx in molecule_idxs:
            nexts = []
            for c in [c1,c2]:
                for c_node in c:
                    nei_mols = mols_from_node(c_node.split()[0])
525
                    nei_mols = [x for x in nei_mols if x > mol_idx]  # fixme: also look at the <= molecules for more robustness
526
527
528
529
530
531
532
533
                    
                    # If there are molecule next
                    if len(nei_mols) > 0:
                        next_nei_mol = min(nei_mols)
                        nexts.append((next_nei_mol))

            nexts.sort(key=lambda x: x)
            quality = sum([1.0/x if mol_idx+x in nexts else 0 for x in range(1,6)]) / sum([1.0/x for x in range(1,6)])
534
            if quality > 0.6:  # eyeballed but still arbitrary
535
                udg_molecules.add(mol_idx)
536
            # print("mol",mol_idx,molecule_idxs,"quality",quality,"nexts",nexts)
537
538
539
540
541
542
       
        udg_molecules_dict[head[0]]=udg_molecules

    # Then: annotate edges as to whether they're real (their udg_molecule(s) are nearby) or fake
    for n1, n2 , data in d2_component.edges(data=True):
        # Parse the current node name
543
        head, c1, c2 = parse_dg_name(d2_component,n1)
544
545
        node_udg_molecules = udg_molecules_dict[head[0]]
        
546
        n_head, n_c1, n_c2 = parse_dg_name(d2_component,n2)
547
548
549
550
551
552
553
        neighbor_udg_molecules = udg_molecules_dict[n_head[0]]
       
        if nearby_udg_molecules(node_udg_molecules, neighbor_udg_molecules):
            color = 'green'
        else:
            color = 'red'
        data['color'] = color
554
        # print("edge",node_udg_molecules,neighbor_udg_molecules,color)
555
556
557
558
    
    # also, annotate nodes by their putative molecule found
    for n, data in d2_component.nodes(data=True):
        # Parse the current node name
559
        head, c1, c2 = parse_dg_name(d2_component,n)
560
561
562
        node_udg_molecules = udg_molecules_dict[head[0]]
        data['udg_molecule']= '_'.join(list(map(str,node_udg_molecules)))
        
563
564
565
566
567
568
569
570
    # aggressive: delete nodes which have either no found udg_molecule, or two udg_molecules
    # turns out it's not a good strategy as the nodes with two udg_molecules are important to connect portions of graph
    # but what if we magically keep those where the two adjacent molecules are close together
    if True:
        d2_component = d2_component.copy()
        nodes_to_remove = []
        for n, data in d2_component.nodes(data=True):
            # Parse the current node name
571
            head, c1, c2 = parse_dg_name(d2_component,n)
572
573
574
            if "_" in data['udg_molecule'] or data['udg_molecule'] == '':
                if "_" in data['udg_molecule']:
                    m1, m2 = list(map(int,data['udg_molecule'].split("_")))
575
                    if abs(m2-m1) < 30: continue  # don't remove that kind of nodes
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
                nodes_to_remove += [n]
        d2_component.remove_nodes_from(nodes_to_remove)
        print("removed",len(nodes_to_remove),"bad nodes")

    # aggressive: delete red edges
    if True:
        d2_component = d2_component.copy()
        edges_to_remove = []
        for n1, n2, data in d2_component.edges(data=True):
            if data['color'] == 'red':
                edges_to_remove += [(n1,n2)]
        d2_component.remove_edges_from(edges_to_remove)
        print("removed",len(edges_to_remove),"bad edges")


    return d2_component
592

593
594
def main():
    args = parse_args()
Yoann Dufresne's avatar
Yoann Dufresne committed
595
    graph = nx.read_gexf(args.filename)
596

Yoann Dufresne's avatar
Yoann Dufresne committed
597
    if args.type == "path":
Yoann Dufresne's avatar
Yoann Dufresne committed
598
599
600
601
        if args.barcode_graph is None:
            print("--barcode_graph is required for path analysis", file=sys.stderr)
            exit(0)

Yoann Dufresne's avatar
Yoann Dufresne committed
602
        barcode_graph = nx.read_gexf(args.barcode_graph)
Yoann Dufresne's avatar
Yoann Dufresne committed
603
604
605
        # if len(list(nx.connected_components(graph))) != 1:
        #     print([len(x) for x in list(nx.connected_components(graph))])
        #     exit("when running evaluate.py --type path, the graph should have a single connected component (it's supposed to be a path, after all)")
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
606
607

        # compute LIS over the path
Yoann Dufresne's avatar
Yoann Dufresne committed
608
609
        # longest_path = compute_longest_increasing_paths(graph)
        longest_path = compute_shortest_edit_path(graph)
Rayan  CHIKHI's avatar
Rayan CHIKHI committed
610
611
        print("--- Largest component analysis ---")
        print(f"Size of the longest path: {len(longest_path)}")
Yoann Dufresne's avatar
Yoann Dufresne committed
612
613
        if not args.light_print:
            print(longest_path)
Yoann Dufresne's avatar
Yoann Dufresne committed
614

Rayan  CHIKHI's avatar
Rayan CHIKHI committed
615
616
617
        # get over/under counted molecules
        print("--- Under/over molecule counts ---")
        frequencies = parse_path_graph_frequencies(graph, barcode_graph)
Yoann Dufresne's avatar
Yoann Dufresne committed
618
        print_path_summary(frequencies, light_print=args.light_print)
Yoann Dufresne's avatar
Yoann Dufresne committed
619
        print(f"Size of the longest path: {len(longest_path)}")
620
621
622
    elif args.type == "dgraphs":
        udg_per_node = parse_udg_qualities(graph)
        # print(udg_per_node)
Yoann Dufresne's avatar
Yoann Dufresne committed
623
    elif args.type == "d2":
Yoann Dufresne's avatar
Yoann Dufresne committed
624

Yoann Dufresne's avatar
Yoann Dufresne committed
625
626
627
628
        components = list(nx.connected_components(graph))
        components.sort(key=lambda x: -len(x))

        component = graph.subgraph(components[0])
629
        longest_path = compute_longest_increasing_paths(component, max_gap=args.max_gap)
Yoann Dufresne's avatar
Yoann Dufresne committed
630
631
        vars = compute_covered_variables(graph, longest_path)
        print_d2_summary(components, longest_path, coverage_vars=vars, light_print=args.light_print)
632
633
634
635
636
637
638
639
640
    
    # added by Rayan
    # example:
    # python evaluate.py --type d2-2annotate ~/Dropbox/cnrs/projects/10x-barcodes/yoann_to_cedric_ilp/d2_graph.gexf
    elif args.type == "d2-2annotate":
        components = list(nx.connected_components(graph))
        components.sort(key=lambda x: -len(x))
        component = graph.subgraph(components[0])
 
641
        component = verify_graph_edges(component)
642

Yoann Dufresne's avatar
Yoann Dufresne committed
643
644
        extension = args.filename.split('.')[-1]
        base_filename = '.'.join(args.filename.split('.')[:-1])
Yoann Dufresne's avatar
Yoann Dufresne committed
645
        nx.write_gexf(component, base_filename+".verified."+extension)
Yoann Dufresne's avatar
Yoann Dufresne committed
646

647
648
649

if __name__ == "__main__":
    main()