Mercurial > repos > yusuf > associate_phenotypes
comparison filter_by_index_gamma @ 0:6411ca16916e default tip
initial commit
author | Yusuf Ali <ali@yusuf.email> |
---|---|
date | Wed, 25 Mar 2015 13:23:29 -0600 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:6411ca16916e |
---|---|
1 #!/usr/bin/env perl | |
2 | |
3 use strict; | |
4 use warnings; | |
5 use DB_File; | |
6 use Parse::BooleanLogic; | |
7 use Math::CDF qw(pgamma qgamma); # relevance score -> gamma p-value | |
8 use PDL qw(pdl); | |
9 use PDL::Stats::Distr qw(mme_gamma); # gamma dist parameter estimates | |
10 use vars qw($parser %cached_sentences %sentence_index); | |
11 | |
12 my $quiet = 0; | |
13 if(@ARGV and $ARGV[0] =~ /^-q/){ | |
14 $quiet = 1; | |
15 shift @ARGV; | |
16 } | |
17 | |
18 @ARGV == 5 or die "Usage: $0 [-q(uiet)] <index filename base> <db name> <hgvs_annotated.txt> <output.txt> <query>\nWhere query has the format \"this or that\", \"this and that\", etc.\n"; | |
19 | |
20 my $signal_p = 0.95; # signal is top 5% of scores | |
21 my $index_filename_base = shift @ARGV; | |
22 my $db_name = shift @ARGV; | |
23 my $hgvs_file = shift @ARGV; | |
24 my $out_file = shift @ARGV; | |
25 my $orig_query = shift @ARGV; | |
26 | |
27 $parser = new Parse::BooleanLogic(operators => ['and', 'or']); | |
28 my $query_tree = $parser->as_array($orig_query, error_cb => sub {die "Could not parse query: @_\n"}); | |
29 # For simplicity, turn the tree into a base set of or statements (which means expanding "A and (B or C)" into "A and B or A and C") a.k.a. "sum of products/minterms" | |
30 my @query_terms = flatten_query($query_tree); | |
31 | |
32 my $df_index_filename = $index_filename_base."df_index"; | |
33 my %df_index; | |
34 my $df_index_handle = tie %df_index, "DB_File", $df_index_filename, O_RDONLY, 0400, $DB_BTREE | |
35 or die "Cannot open $df_index_filename: $!\n"; | |
36 my $gene_record_count = $df_index{"__DOC_COUNT__"}; | |
37 | |
38 my $sentence_index_filename = $index_filename_base."sentence_index"; | |
39 my $sentence_index_handle = tie %sentence_index, "DB_File", $sentence_index_filename, O_RDONLY, 0400, $DB_HASH | |
40 or die "Cannot open $sentence_index_filename: $!\n"; | |
41 | |
42 # Get the list of gene symbols we'll need | |
43 open(HGVS, $hgvs_file) | |
44 or die "Cannot open $hgvs_file for reading: $!\n"; | |
45 my $header = <HGVS>; | |
46 chomp $header; | |
47 my @header_columns = split /\t/, $header; | |
48 my ($gene_name_column, $chr_column, $from_column, $to_column); | |
49 for(my $i = 0; $i <= $#header_columns; $i++){ | |
50 if($header_columns[$i] eq "Gene Name"){ | |
51 $gene_name_column = $i; | |
52 } | |
53 elsif($header_columns[$i] eq "Chr"){ | |
54 $chr_column = $i; | |
55 } | |
56 elsif($header_columns[$i] eq "DNA From"){ | |
57 $from_column = $i; | |
58 } | |
59 elsif($header_columns[$i] eq "DNA To"){ | |
60 $to_column = $i; | |
61 } | |
62 } | |
63 my $blank_query = not @query_terms; | |
64 # Special case of empty query means print all info for variant ranges listed in the input HGVS file (assuming the DB was indexed to include chr:pos keys) | |
65 if($blank_query){ | |
66 #print STDERR "Running blank query\n"; | |
67 if(not defined $chr_column){ | |
68 die "Could not find 'Chr' column in the input header, aborting\n"; | |
69 } | |
70 if(not defined $from_column){ | |
71 die "Could not find 'DNA From' column in the input header, aborting\n"; | |
72 } | |
73 if(not defined $to_column){ | |
74 die "Could not find 'DNA To' column in the input header, aborting\n"; | |
75 } | |
76 # Build the list of locations that will need to be searched in the index | |
77 | |
78 open(OUT, ">$out_file") | |
79 or die "Cannot open $out_file for writing: $!\n"; | |
80 print OUT $header, "\t$db_name Text Matches\n"; | |
81 | |
82 while(<HGVS>){ | |
83 chomp; | |
84 my @F = split /\t/, $_, -1; | |
85 my @pos_data; | |
86 for my $pos ($F[$from_column]..$F[$to_column]){ # for each position in the range | |
87 my $pos_match_data = fetch_sentence("$F[$chr_column]:$pos", -1); # fetch all data for this position | |
88 push @pos_data, "*$F[$chr_column]:$pos* ".$pos_match_data if defined $pos_match_data; | |
89 } | |
90 print OUT join("\t", @F, join(" // ", @pos_data)),"\n"; | |
91 } | |
92 close(OUT); | |
93 exit; | |
94 } | |
95 elsif(not defined $gene_name_column){ | |
96 die "Could not find 'Gene Name' column in the input header, aborting\n"; | |
97 } | |
98 #print STDERR "Query terms: " , scalar(@query_terms), "\n"; | |
99 my %gene_to_query_match_ranges; | |
100 # Determine the set of genes that might match the query, based on the word index | |
101 for my $query_term (@query_terms){ | |
102 #print STDERR "Query term $query_term\n"; | |
103 my %doc_hits; # how many needed words match the document? | |
104 my $contiguous = 1; #by default multiword queries must be contiguous | |
105 # Unless it's an AND query | |
106 if($query_term =~ s/ and / /g){ | |
107 $contiguous = 0; | |
108 } | |
109 | |
110 my @words = split /\s+/, $query_term; # can be multi-word term like "mental retardation" | |
111 for(my $i = 0; $i <= $#words; $i++){ | |
112 my $word = mc($words[$i]); # can be a stem word, like hypoton | |
113 #print STDERR "Checking word $word..."; | |
114 if($i == 0){ | |
115 my $first_word_docs = get_doc_offsets($df_index_handle, $word); # get all words' docs off this stem | |
116 #print STDERR scalar(keys %$first_word_docs), " documents found\n"; | |
117 for my $doc (keys %$first_word_docs){ | |
118 $doc_hits{$doc} = $first_word_docs->{$doc}; # populate initial hit list that'll be whittled down in subsequent outer loops of multiword phrase members | |
119 } | |
120 next; | |
121 } | |
122 my @candidate_docs = keys %doc_hits; | |
123 last if not @candidate_docs; # short circuit searches guaranteed to fail | |
124 | |
125 # each additional word must directly follow an existing match | |
126 my $word_doc_offsets_ref = get_doc_offsets($df_index_handle, $word); # get all words' docs off this stem | |
127 #print STDERR scalar(keys %$word_doc_offsets_ref), " documents found\n"; | |
128 for my $doc (@candidate_docs){ | |
129 my $num_matches = 0; | |
130 if(not exists $word_doc_offsets_ref->{$doc}){ # required word missing, eliminate doc from consideration | |
131 delete $doc_hits{$doc}; | |
132 next; | |
133 } | |
134 # see if any of the instances of the additional words directly follow the last word we successfully matched | |
135 my $so_far_matches_ref = $doc_hits{$doc}; | |
136 my $next_word_matches_ref = $word_doc_offsets_ref->{$doc}; | |
137 for (my $j=0; $j <= $#{$so_far_matches_ref}; $j++){ | |
138 my $existing_match_extended = 0; | |
139 next unless defined $so_far_matches_ref->[$j]->[2]; # every once in a while there is no article id parsed | |
140 for (my $k=0; $k <= $#{$next_word_matches_ref}; $k++){ | |
141 # Same article? | |
142 next unless defined $next_word_matches_ref->[$k]->[2] and $next_word_matches_ref->[$k]->[2] eq $so_far_matches_ref->[$j]->[2]; | |
143 if(not $contiguous){ | |
144 $so_far_matches_ref->[$j]->[4] .= " AND ".$next_word_matches_ref->[$k]->[4]; # update the matched term to include the extension too | |
145 if(ref $so_far_matches_ref->[$j]->[3] ne "ARRAY"){ # match does not yet span multiple sentences | |
146 last if $next_word_matches_ref->[$k]->[3] == $so_far_matches_ref->[$j]->[3]; # same sentence | |
147 $so_far_matches_ref->[$j]->[3] = [$so_far_matches_ref->[$j]->[3], $next_word_matches_ref->[$k]->[3]]; # change from scalar to array (of sentence numbers) | |
148 } | |
149 elsif(not grep {$_ eq $next_word_matches_ref->[$k]->[3]} @{$so_far_matches_ref->[$j]->[3]}){ | |
150 push @{$so_far_matches_ref->[$j]->[3]}, $next_word_matches_ref->[$k]->[3]; # add top spanning sentences list of not already there | |
151 } | |
152 } | |
153 # else contiguous word occurences required. | |
154 # Same sentence? | |
155 next unless $next_word_matches_ref->[$k]->[3] == $so_far_matches_ref->[$j]->[3]; | |
156 | |
157 my $space_between_match_words = $next_word_matches_ref->[$k]->[0] - $so_far_matches_ref->[$j]->[1]; | |
158 if($space_between_match_words <= 2){ | |
159 $existing_match_extended = 1; | |
160 $so_far_matches_ref->[$j]->[1] = $next_word_matches_ref->[$k]->[1]; # move the match cursor to include the new extending word | |
161 $so_far_matches_ref->[$j]->[4] .= " ".$next_word_matches_ref->[$k]->[4]; # update the matched term to include the extension too | |
162 last; | |
163 } | |
164 elsif($space_between_match_words > 2){ # more than two typographical symbols between words, consider non-continuous | |
165 last; # since the offsets are in order, any further k would only yield a larger spacing, so shortcircuit | |
166 } | |
167 } | |
168 if(not $existing_match_extended){ | |
169 splice(@$so_far_matches_ref, $j, 1); | |
170 $j--; | |
171 } | |
172 else{ | |
173 $num_matches++; | |
174 } | |
175 } | |
176 if(not $num_matches){ | |
177 delete $doc_hits{$doc}; | |
178 } | |
179 } | |
180 } | |
181 # the only keys that get to this point should be those that match all terms | |
182 for my $doc (keys %doc_hits){ | |
183 $gene_to_query_match_ranges{$doc} = [] if not exists $gene_to_query_match_ranges{$doc}; | |
184 push @{$gene_to_query_match_ranges{$doc}}, [$query_term, @{$doc_hits{$doc}}]; | |
185 } | |
186 } | |
187 | |
188 my @matched_genes = keys %gene_to_query_match_ranges; | |
189 #print STDERR "Found ", scalar(@matched_genes), "/$gene_record_count records in cached iHOP matching the query\n" unless $quiet; | |
190 my %query_gene_counts; | |
191 my %ntf; | |
192 for my $gene (keys %gene_to_query_match_ranges){ | |
193 my $max_doc_word_count = $df_index{"__DOC_MAX_WC_$gene"}; | |
194 for my $count_record (@{$gene_to_query_match_ranges{$gene}}){ | |
195 my ($query_term, @query_term_match_ranges_in_this_gene) = @$count_record; | |
196 # next if $query_term eq $gene; # slightly controversial? exclude references to genes from the score if the gene is the record being talked about (obviously it will be highly scored) | |
197 # allows us to find first degree interactors (i.e. points for "A interacts with B", in the record describing A) without creating crazy high score for doc describing gene B if B was in the original query without any phenotype query terms | |
198 $query_gene_counts{$query_term}++; | |
199 | |
200 $ntf{$gene} = {} unless exists $ntf{$gene}; | |
201 # atypical use of log in order to weigh heavy use of a common term less than occasional use of a rare term | |
202 $ntf{$gene}->{$query_term} = log($#query_term_match_ranges_in_this_gene+2)/log($max_doc_word_count+1); | |
203 } | |
204 #print STDERR "Doc max word count is $max_doc_word_count for $gene, ntf keys = ", keys %{$ntf{$gene}}, "\n"; | |
205 } | |
206 | |
207 my %idf; | |
208 for my $query_term (@query_terms){ # convert %idf values from documents-with-the-query-term-count to actual IDF | |
209 next unless exists $query_gene_counts{$query_term}; # query not in the document collection | |
210 $idf{$query_term} = log($gene_record_count/$query_gene_counts{$query_term}); | |
211 #print STDERR "$query_term IDF is $idf{$query_term}\n"; | |
212 } | |
213 | |
214 # Create a relevance score using a normalized term frequency - inverse document frequency summation | |
215 my %relevance_score; | |
216 my %matched_query_terms; | |
217 for my $gene_symbol (keys %gene_to_query_match_ranges){ | |
218 my $relevance_score = 0; | |
219 # Hmm, take average, sum or max of TF-IDFs? | |
220 my $max_query_score = 0; | |
221 my @matched_query_terms; | |
222 my $query_score = 0; | |
223 for (my $i = 0; $i <= $#query_terms; $i++){ | |
224 my $query_term = $query_terms[$i]; | |
225 next unless exists $idf{$query_term}; | |
226 next unless exists $ntf{$gene_symbol}->{$query_term}; | |
227 $query_score += $ntf{$gene_symbol}->{$query_term}*$idf{$query_term}; | |
228 push @matched_query_terms, $query_term; | |
229 $query_score *= 1-$i/scalar(@query_terms)/2 if scalar(@query_terms) > 2;# adjust the query score so the first terms are weighted more heavily if a bunch of terms are being searched | |
230 $max_query_score = $query_score if $query_score > $max_query_score; | |
231 $relevance_score += $query_score; | |
232 } | |
233 # this square root trick will not affect the score of a single term query, but will penalize a high total score that is comprised of a bunch of low value individual term scores) | |
234 $relevance_score{$gene_symbol} = sqrt($relevance_score*$max_query_score); | |
235 #print STDERR "Relevance score for $gene_symbol is $relevance_score{$gene_symbol}\n"; | |
236 $matched_query_terms{$gene_symbol} = \@matched_query_terms; | |
237 } | |
238 | |
239 # Characterize relevance score as a gamma statistical distribution and convert to probability | |
240 my $max_relevance_score = 0; | |
241 for my $relevance_score (values %relevance_score){ | |
242 $max_relevance_score = $relevance_score if $relevance_score > $max_relevance_score; | |
243 } | |
244 # Remove top end scores as signal, characterize the rest as noise. | |
245 # Iterative estimation of gamma parameters and removing data within range where CDF>99% | |
246 my $noise_data = pdl(values %relevance_score); | |
247 my ($shape, $scale) = $noise_data->mme_gamma(); | |
248 #print STDERR "Initial gamma distribution estimates: $shape, $scale (max observation $max_relevance_score)\n"; | |
249 my $signal_cutoff = qgamma($signal_p, $shape, 1/$scale); | |
250 my @noise_data; | |
251 for my $gene_symbol (keys %relevance_score){ | |
252 my $score = $relevance_score{$gene_symbol}; | |
253 push @noise_data, $score if $score < $signal_cutoff; | |
254 } | |
255 $noise_data = pdl(@noise_data); | |
256 ($shape, $scale) = $noise_data->mme_gamma(); | |
257 #print STDERR "Revised gamma distribution estimates (noise estimate at $signal_cutoff (CDF $signal_p)): $shape, $scale\n"; | |
258 # Convert scores to probabilities | |
259 for my $gene_symbol (keys %relevance_score){ | |
260 $relevance_score{$gene_symbol} = 1-pgamma($relevance_score{$gene_symbol}, $shape, 1/$scale); | |
261 } | |
262 | |
263 #TODO: create summary stats for each query term so the user gets an idea of each's contribution? | |
264 | |
265 my %pubmed_matches; | |
266 for my $gene_symbol (keys %gene_to_query_match_ranges){ | |
267 my $query_match_ranges_ref = $gene_to_query_match_ranges{$gene_symbol}; | |
268 my %matching_sentences; | |
269 for my $count_record (@$query_match_ranges_ref){ | |
270 my ($query_term, @query_term_match_ranges_in_this_gene) = @$count_record; | |
271 for my $occ_info (@query_term_match_ranges_in_this_gene){ | |
272 my $id = $occ_info->[2]; | |
273 my $sentence_number = $occ_info->[3]; | |
274 my $query_match_word = $occ_info->[4]; | |
275 # Fetch the preparsed sentence from the sentence index based on id and sentence number | |
276 # Will automatically *HIGHLIGHT* the query terms fetched in the sentence over the course of this script | |
277 if(ref $sentence_number eq "ARRAY"){ # match spans multiple sentences | |
278 for my $s (@$sentence_number){ | |
279 for my $word (split / AND /, $query_match_word){ | |
280 #print STDERR "Highlighting $word in $id #$s for query term $query_term (multisentence match)\n"; | |
281 $matching_sentences{fetch_sentence_key($id, $s, $word)}++; | |
282 } | |
283 } | |
284 } | |
285 else{ # single sentence match | |
286 #print STDERR "Highlighting $query_match_word in $id #$sentence_number for query term $query_term\n"; | |
287 $matching_sentences{fetch_sentence_key($id, $sentence_number, $query_match_word)}++; | |
288 } | |
289 } | |
290 } | |
291 $gene_symbol =~ s/_/\//; # didn't have a forward slash in a gene name for disk caching purposes | |
292 if(keys %matching_sentences){ | |
293 $pubmed_matches{$gene_symbol} = [] unless exists $pubmed_matches{$gene_symbol}; | |
294 for my $new_match_ref (keys %matching_sentences){ | |
295 push @{$pubmed_matches{$gene_symbol}}, $new_match_ref unless grep {$_ eq $new_match_ref} @{$pubmed_matches{$gene_symbol}}; # only put in new sentences, no need to dup | |
296 } | |
297 } | |
298 } | |
299 | |
300 $orig_query =~ s/\s+/ /; # normalized whitespace | |
301 $orig_query =~ s/ and / and /i; # lc() | |
302 my @orig_query_terms = split /\s+or\s+/, $orig_query; | |
303 | |
304 open(OUT, ">$out_file") | |
305 or die "Cannot open $out_file for writing: $!\n"; | |
306 my $new_header = $header; | |
307 $new_header .= "\t$db_name p-value (log normalized TF-IDF score, gamma dist)\t$db_name Matching Terms ($orig_query)\t$db_name Text Matches"; | |
308 print OUT $new_header, "\n"; | |
309 | |
310 # Check if any of the variants in the annotated HGVS table are in genes from the OMIM match list | |
311 while(<HGVS>){ | |
312 chomp; | |
313 my @F = split /\t/, $_, -1; | |
314 # order the ids from highest number of sentence matches to lowest, from highest ranked term to least | |
315 my (%id2match_count, %id2sentences); | |
316 my @matched_genes; | |
317 my $relevance_score_final = 1; | |
318 my @matched_query_terms; | |
319 for my $gene_name (split /\s*;\s*/, $F[$gene_name_column]){ | |
320 next unless exists $pubmed_matches{$gene_name}; | |
321 push @matched_genes, $gene_name; | |
322 for my $sentence_ref (@{$pubmed_matches{$gene_name}}){ # 0 == always fetch the title which is stored in sentence index 0 | |
323 my $pubmed_record = fetch_sentence($sentence_ref); | |
324 $id2match_count{$pubmed_record->[0]}++; # key = id | |
325 if(not exists $id2sentences{$pubmed_record->[0]}){ | |
326 $id2sentences{$pubmed_record->[0]} = {}; | |
327 my $title_record = fetch_sentence(fetch_sentence_key($pubmed_record->[0], 0, "")); | |
328 next unless $title_record->[0]; | |
329 print STDERR "No $index_filename_base sentence number for ", $title_record->[0], "\n" if not defined $title_record->[1]; | |
330 print STDERR "No $index_filename_base sentence text for ", $title_record->[0], " sentence #", $title_record->[1], "\n" if not defined $title_record->[2]; | |
331 $id2sentences{$title_record->[0]}->{$title_record->[2]} = $title_record->[1]; | |
332 } | |
333 # Only print sentences that match a query term other than the gene name for the record key, if that gene name is part of the query | |
334 my $non_self_query_ref = 0; | |
335 while($pubmed_record->[2] =~ /\*(.+?)\*/g){ | |
336 if($1 ne $gene_name){ | |
337 $non_self_query_ref = 1; | |
338 last; | |
339 } | |
340 } | |
341 #print STDERR "rejected $gene_name self-only sentence ",$pubmed_record->[2],"\n" unless $non_self_query_ref; | |
342 next unless $non_self_query_ref; | |
343 $id2sentences{$pubmed_record->[0]}->{$pubmed_record->[2]} = $pubmed_record->[1]; # value = sentence order within pubmed text | |
344 } | |
345 $relevance_score_final *= $relevance_score{$gene_name}; | |
346 push @matched_query_terms, @{$matched_query_terms{$gene_name}}; | |
347 } | |
348 | |
349 # If we get here, there were matches | |
350 my @ordered_ids = sort {$id2match_count{$b} <=> $id2match_count{$a}} keys %id2match_count; | |
351 | |
352 # print sentences in each id in order, with ellipsis if not contiguous | |
353 my %h; | |
354 print OUT join("\t", @F, ($relevance_score_final != 1 ? $relevance_score_final : ""), (@matched_query_terms ? join("; ", sort grep {not $h{$_}++} @matched_query_terms) : "")), "\t"; | |
355 my $first_record = 1; | |
356 for my $id (@ordered_ids){ | |
357 my $sentence2order = $id2sentences{$id}; | |
358 my @ordered_sentences = sort {$sentence2order->{$a} <=> $sentence2order->{$b}} keys %$sentence2order; | |
359 next if scalar(@ordered_sentences) == 1; # due to self-gene only referencing filter above, we may have no matching sentences in a record. Skip in this case. | |
360 if($first_record){ | |
361 $first_record = 0; | |
362 } | |
363 else{ | |
364 print OUT " // "; | |
365 } | |
366 my $title = shift(@ordered_sentences); | |
367 print OUT "$db_name $id",(defined $title ? " $title": ""),":"; # first sentence is always the record title | |
368 my $last_ordinal = 0; | |
369 for my $s (@ordered_sentences){ | |
370 if($last_ordinal and $sentence2order->{$s} != $last_ordinal+1){ | |
371 print OUT ".."; | |
372 } | |
373 print OUT " ",$s; | |
374 $last_ordinal = $sentence2order->{$s}; | |
375 } | |
376 } | |
377 print OUT "\n"; | |
378 } | |
379 | |
380 sub get_doc_offsets{ | |
381 my ($db_handle, $word_stem) = @_; | |
382 my %doc2offsets; | |
383 | |
384 my $is_uc = $word_stem =~ /^[A-Z0-9]+$/; | |
385 my $has_wildcard = $word_stem =~ s/\*$//; | |
386 my $value = 0; | |
387 my $cursor_key = $word_stem; | |
388 # retrieves the first | |
389 for(my $status = $db_handle->seq($cursor_key, $value, R_CURSOR); | |
390 $status == 0; | |
391 $status = $db_handle->seq($cursor_key, $value, R_NEXT)){ | |
392 if(CORE::index($cursor_key,$word_stem) != 0){ | |
393 last; # outside the records that have the requested stem now | |
394 } | |
395 for my $record (split /\n/s, $value){ | |
396 my ($doc, @occ_infos) = split /:/, $record; | |
397 $doc2offsets{$doc} = [] if not exists $doc2offsets{$doc}; | |
398 for my $occ_info (@occ_infos){ | |
399 my ($term_offset, $id, $sentence_number) = split /,/, $occ_info, -1; | |
400 # record start and end of word to facilitate partial key consecutive word matching algorithm used in this script | |
401 push @{$doc2offsets{$doc}}, [$term_offset, $term_offset+length($cursor_key), $id, $sentence_number, $cursor_key]; | |
402 } | |
403 } | |
404 last if $is_uc and not $has_wildcard; # only exact matches for upper case words like gene names | |
405 } | |
406 return \%doc2offsets; | |
407 } | |
408 | |
409 sub mc{ | |
410 if($_[0] =~ /^[A-Z][a-z]+$/){ | |
411 return lc($_[0]); # sentence case normalization to lower case for regular words | |
412 } | |
413 else{ | |
414 return $_[0]; # as-is for gene names, etc | |
415 } | |
416 } | |
417 | |
418 sub fetch_sentence_key{ | |
419 my ($id, $sentence_number, $query_term) = @_; | |
420 | |
421 $sentence_number = 0 if not defined $sentence_number; | |
422 return ":$sentence_number" if not $id; | |
423 my $key = "$id:$sentence_number"; | |
424 if(not exists $cached_sentences{$key}){ | |
425 my @sentences = split /\n/, $sentence_index{$id}; | |
426 $cached_sentences{$key} = $sentences[$sentence_number]; | |
427 } | |
428 $cached_sentences{$key} =~ s/\b\Q$query_term\E\b(?!\*)/"*".uc($query_term)."*"/ge unless $query_term eq ""; | |
429 #print STDERR "Highlighted $query_term in $cached_sentences{$key}\n" if $query_term =~ /cirrhosis/; | |
430 return $key; | |
431 } | |
432 | |
433 sub fetch_sentence{ | |
434 if(@_ == 1){ # from cache | |
435 return [split(/:/, $_[0]), $cached_sentences{$_[0]}]; | |
436 } | |
437 else{ # if more than one arg, DIRECT FROM index key as first arg, sentence # is second arg | |
438 return undef if not exists $sentence_index{$_[0]}; | |
439 my @sentences = split /\n/, $sentence_index{$_[0]}; | |
440 if($_[1] < 0){ # all sentences request | |
441 return join("; ", @sentences); | |
442 } | |
443 return $sentences[$_[1]]; | |
444 } | |
445 } | |
446 | |
447 | |
448 # boolean operator tree to flat expanded single depth "or" op query | |
449 sub flatten_query{ | |
450 my $tree = shift @_; | |
451 my @or_queries; | |
452 | |
453 # Base case: the tree is just a leaf (denoted by a hash reference). Return value of the operand it represents. | |
454 if(ref $tree eq "HASH"){ | |
455 return ($tree->{"operand"}); | |
456 } | |
457 | |
458 elsif(not ref $tree){ | |
459 return $tree; | |
460 } | |
461 | |
462 # Otherwise it's an operation array | |
463 if(ref $tree ne "ARRAY"){ | |
464 die "Could not parse $tree, logic error in the query parser\n"; | |
465 } | |
466 | |
467 # Deal with AND first since it has higher precedence | |
468 for (my $i = 1; $i < $#{$tree}; $i++){ | |
469 if($tree->[$i] eq "and"){ | |
470 my @expanded_term; | |
471 my @t1_terms = flatten_query($tree->[$i-1]); | |
472 my @t2_terms = flatten_query($tree->[$i+1]); | |
473 #print STDERR "need to expand ", $tree->[$i-1], "(@t1_terms) AND ", $tree->[$i+1], "(@t2_terms)\n"; | |
474 for my $term1 (@t1_terms){ | |
475 for my $term2 (@t2_terms){ | |
476 #print STDERR "Expanding to $term1 and $term2\n"; | |
477 push @expanded_term, "$term1 and $term2"; | |
478 } | |
479 } | |
480 splice(@$tree, $i-1, 3, @expanded_term); | |
481 $i--; # list has been shortened | |
482 } | |
483 } | |
484 # Should be only "OR" ops left | |
485 # Resolve any OR subtrees | |
486 for(my $i = 0; $i <= $#{$tree}; $i++){ | |
487 next if $tree->[$i] eq "or"; | |
488 push @or_queries, flatten_query($tree->[$i]); # otherwise recursive parse | |
489 } | |
490 | |
491 return @or_queries; | |
492 } |