1 /*****************************************************************
2 * SQUID - a library of functions for biological sequence analysis
3 * Copyright (C) 1992-2002 Washington University School of Medicine
5 * This source code is freely distributed under the terms of the
6 * GNU General Public License. See the files COPYRIGHT and LICENSE
8 *****************************************************************/
12 * Comparison of multiple alignments. Three functions are
13 * provided, using subtly different scoring schemes:
14 * CompareMultAlignments() - basic scoring scheme
15 * CompareRefMultAlignments() - only certain "canonical" columns
18 * The similarity measure is a fractional alignment identity averaged
19 * over all sequence pairs. The score for all pairs is:
20 * (identically aligned symbols) / (total aligned columns in
23 * A column c is identically aligned for sequences i, j if:
24 * 1) both i,j have a symbol aligned in column c, and the
25 * same pair of symbols is aligned somewhere in the test
27 * 2) S[i][c] is aligned to a gap in sequence j, and that symbol
28 * is aligned to a gap in the test alignment
32 * The algorithm is as follows:
33 * 1) For each known/test aligned pair of sequences (k1,k2 and t1,t2)
34 * construct a list for each sequence, in which for every
35 * counted symbol we record the raw index of the symbol in
36 * the other sequence that it aligns to, or -1 if it aligns
37 * to a gap or uncounted symbol.
39 * 2) Compare the list for k1 to the list for t1 and count an identity
40 * for each correct alignment.
42 * 3) Repeat 2) for comparing k2 to t2. Note that this means correct sym/sym
43 * alignments count for 2; correct sym/gap alignments count for 1.
45 * 4) The score is (identities from 2 + identities from 3) /
46 * (totals from 2 + totals from 3).
48 * Written originally for koala's ss2 pairwise alignment package.
50 * Sean Eddy, Sun Nov 1 12:45:11 1992
51 * SRE, Thu Jul 29 16:47:18 1993: major revision: all functions replaced by new algorithm
52 * CVS $Id: aligneval.c,v 1.7 2002/10/09 14:26:09 eddy Exp)
60 #include "sre_random.h"
62 static int make_alilist(char *s1, char *s2, int **ret_s1_list, int *ret_listlen);
63 static int make_ref_alilist(int *refcoords, char *k1, char *k2, char *s1, char *s2,
64 int **ret_s1_list, int *ret_listlen);
65 static int compare_lists(int *k1, int *k2, int *t1, int *t2, int len1, int len2, float *ret_sc);
68 /* Function: ComparePairAlignments
70 * Purpose: Calculate and return a number representing how well two different alignments
71 * of a pair of sequences compare. The number is, roughly speaking,
72 * the fraction of columns which are identically aligned.
74 * For all columns c in which either known1[c] or known2[c]
75 * is a non-gap, count an identity if those same symbols are
76 * aligned somewhere in calc1/calc2. The score is identities/total
77 * columns examined. (i.e. fully gapped columns don't count)
79 * more explicitly, identities come from:
80 * both known and test aligned pairs have the same symbol in the first sequence aligned to
81 * a gap in the second sequence;
82 * both known and test aligned pairs have the same symbol in the second sequence
83 * aligned to a gap in the first sequence;
84 * the known alignment has symbols aligned at this column, and the test
85 * alignment aligns the same two symbols.
87 * Args: known1, known2: trusted alignment of two sequences
88 * calc1, calc2: test alignment of two sequences
90 * Return: Returns -1.0 on internal failure.
93 ComparePairAlignments(char *known1, char *known2, char *calc1, char *calc2)
102 if (! make_alilist(calc1, calc2, &tlist1, &len1)) return -1.0;
103 if (! make_alilist(calc2, calc1, &tlist2, &len2)) return -1.0;
104 if (! make_alilist(known1, known2, &klist1, &len1)) return -1.0;
105 if (! make_alilist(known2, known1, &klist2, &len2)) return -1.0;
106 if (! compare_lists(klist1, klist2, tlist1, tlist2, len1, len2, &score)) return -1.0;
117 /* Function: CompareRefPairAlignments()
119 * Same as above, but the only columns that count are the ones
120 * with indices in *refcoord. *refcoord and the known1, known2
121 * pair must be in sync with each other (come from the same
122 * multiple sequence alignment)
124 * Args: ref - 0..alen-1 array of 1 or 0
125 * known1,known2 - trusted alignment
126 * calc1, calc2 - test alignment
128 * Return: the fractional alignment identity on success, -1.0 on failure.
131 CompareRefPairAlignments(int *ref, char *known1, char *known2, char *calc1, char *calc2)
140 if (! make_ref_alilist(ref, known1, known2, calc1, calc2, &tlist1, &len1)) return -1.0;
141 if (! make_ref_alilist(ref, known2, known1, calc2, calc1, &tlist2, &len2)) return -1.0;
142 if (! make_ref_alilist(ref, known1, known2, known1, known2, &klist1, &len1)) return -1.0;
143 if (! make_ref_alilist(ref, known2, known1, known2, known1, &klist2, &len2)) return -1.0;
144 if (! compare_lists(klist1, klist2, tlist1, tlist2, len1, len2, &score)) return -1.0;
153 /* Function: make_alilist()
155 * Purpose: Construct a list (array) mapping the raw symbols of s1
156 * onto the indexes of the aligned symbols in s2 (or -1
157 * for gaps in s2). The list (s1_list) will be of the
158 * length of s1's raw sequence.
160 * Args: s1 - sequence to construct the list for
161 * s2 - sequence s1 is aligned to
162 * ret_s1_list - RETURN: the constructed list (caller must free)
163 * ret_listlen - RETURN: length of the list
165 * Returns: 1 on success, 0 on failure
168 make_alilist(char *s1, char *s2, int **ret_s1_list, int *ret_listlen)
171 int col; /* column position in alignment */
172 int r1, r2; /* raw symbol index at current col in s1, s2 */
174 /* Malloc for s1_list. It can't be longer than s1 itself; we just malloc
175 * for that (and waste a wee bit of space)
177 s1_list = (int *) MallocOrDie (sizeof(int) * strlen(s1));
179 for (col = 0; s1[col] != '\0'; col++)
181 /* symbol in s1? Record what it's aligned to, and bump
184 if (! isgap(s1[col]))
186 s1_list[r1] = isgap(s2[col]) ? -1 : r2;
190 /* symbol in s2? bump the r2 counter
192 if (! isgap(s2[col]))
197 *ret_s1_list = s1_list;
203 /* Function: make_ref_alilist()
205 * Purpose: Construct a list (array) mapping the raw symbols of s1
206 * which are under canonical columns of the ref alignment
207 * onto the indexes of the aligned symbols in s2 (or -1
208 * for gaps in s2 or noncanonical symbols in s2).
210 * Args: ref: - array of indices of canonical coords (1 canonical, 0 non)
211 * k1 - s1's known alignment (w/ respect to refcoords)
212 * k2 - s2's known alignment (w/ respect to refcoords)
213 * s1 - sequence to construct the list for
214 * s2 - sequence s1 is aligned to
215 * ret_s1_list - RETURN: the constructed list (caller must free)
216 * ret_listlen - RETURN: length of the list
218 * Returns: 1 on success, 0 on failure
222 make_ref_alilist(int *ref, char *k1, char *k2,
223 char *s1, char *s2, int **ret_s1_list, int *ret_listlen)
226 int col; /* column position in alignment */
227 int r1, r2; /* raw symbol index at current col in s1, s2 */
228 int *canons1; /* flag array, 1 if position i in s1 raw seq is canonical */
229 int lpos; /* position in list */
231 /* Allocations. No arrays can exceed the length of their
232 * appropriate parent (s1 or s2)
234 s1_list = (int *) MallocOrDie (sizeof(int) * strlen(s1));
235 canons1 = (int *) MallocOrDie (sizeof(int) * strlen(s1));
237 /* First we use refcoords and k1,k2 to construct an array of 1's
238 * and 0's, telling us whether s1's raw symbol number i is countable.
239 * It's countable simply if it's under a canonical column.
242 for (col = 0; k1[col] != '\0'; col++)
244 if (! isgap(k1[col]))
246 canons1[r1] = ref[col] ? 1 : 0;
251 /* Now we can construct the list. We don't count pairs if the sym in s1
253 * We have to keep separate track of our position in the list (lpos)
254 * from our positions in the raw sequences (r1,r2)
257 for (col = 0; s1[col] != '\0'; col++)
259 if (! isgap(s1[col]) && canons1[r1])
261 s1_list[lpos] = isgap(s2[col]) ? -1 : r2;
265 if (! isgap(s1[col]))
267 if (! isgap(s2[col]))
273 *ret_s1_list = s1_list;
277 /* Function: compare_lists()
279 * Purpose: Given four alignment lists (k1,k2, t1,t2), calculate the
282 * Args: k1 - list of k1's alignment to k2
283 * k2 - list of k2's alignment to k1
284 * t1 - list of t1's alignment to t2
285 * t2 - list of t2's alignment to t2
286 * len1 - length of k1, t1 lists (same by definition)
287 * len2 - length of k2, t2 lists (same by definition)
288 * ret_sc - RETURN: identity score of alignment
290 * Return: 1 on success, 0 on failure.
293 compare_lists(int *k1, int *k2, int *t1, int *t2, int len1, int len2, float *ret_sc)
300 for (i = 0; i < len1; i++)
303 if (t1[i] == k1[i]) id += 1.0;
306 for ( i = 0; i < len2; i++)
309 if (k2[i] == t2[i]) id += 1.0;
317 /* Function: CompareMultAlignments
319 * Purpose: Invokes pairwise alignment comparison for every possible pair,
320 * and returns the average score over all N(N-1) of them or -1.0
321 * on an internal failure.
323 * Can be slow for large N, since it's quadratic.
325 * Args: kseqs - trusted multiple alignment
326 * tseqs - test multiple alignment
327 * N - number of sequences
329 * Return: average identity score, or -1.0 on failure.
332 CompareMultAlignments(char **kseqs, char **tseqs, int N)
334 int i, j; /* counters for sequences */
336 float tot_score = 0.0;
337 /* do all pairwise comparisons */
338 for (i = 0; i < N; i++)
339 for (j = i+1; j < N; j++)
341 score = ComparePairAlignments(kseqs[i], kseqs[j], tseqs[i], tseqs[j]);
342 if (score < 0.0) return -1.0;
345 return ((tot_score * 2.0) / ((float) N * ((float) N - 1.0)));
350 /* Function: CompareRefMultAlignments()
352 * Purpose: Same as above, except an array of reference coords for
353 * the canonical positions of the known alignment is also
356 * Args: ref : 0..alen-1 array of 1/0 flags, 1 if canon
357 * kseqs : trusted alignment
358 * tseqs : test alignment
359 * N : number of sequences
361 * Return: average identity score, or -1.0 on failure
364 CompareRefMultAlignments(int *ref, char **kseqs, char **tseqs, int N)
366 int i, j; /* counters for sequences */
368 float tot_score = 0.0;
370 /* do all pairwise comparisons */
371 for (i = 0; i < N; i++)
372 for (j = i+1; j < N; j++)
374 score = CompareRefPairAlignments(ref, kseqs[i], kseqs[j], tseqs[i], tseqs[j]);
375 if (score < 0.0) return -1.0;
378 return ((tot_score * 2.0)/ ((float) N * ((float) N - 1.0)));
381 /* Function: PairwiseIdentity()
383 * Purpose: Calculate the pairwise fractional identity between
384 * two aligned sequences s1 and s2. This is simply
385 * (idents / MIN(len1, len2)).
387 * Note how many ways there are to calculate pairwise identity,
388 * because of the variety of choices for the denominator:
389 * idents/(idents+mismat) has the disadvantage that artifactual
390 * gappy alignments would have high "identities".
391 * idents/(AVG|MAX)(len1,len2) both have the disadvantage that
392 * alignments of fragments to longer sequences would have
393 * artifactually low "identities".
395 * Original Case sensitive; also, watch out in nucleic acid alignments;
396 * U/T RNA/DNA alignments will be counted as mismatches!
398 * Clustal Omega patch: Case insensitive and T and U are treated the same
401 PairwiseIdentity(char *s1, char *s2)
403 int idents; /* total identical positions */
404 int len1, len2; /* lengths of seqs */
405 int x; /* position in aligned seqs */
407 idents = len1 = len2 = 0;
408 for (x = 0; s1[x] != '\0' && s2[x] != '\0'; x++)
411 char c1 = toupper(s1[x]);
412 char c2 = toupper(s2[x]);
428 if (s1[x] == s2[x]) idents++;
430 if (!isgap(s2[x])) len2++;
433 if (len2 < len1) len1 = len2;
434 return (len1 == 0 ? 0.0 : (float) idents / (float) len1);
439 /* Function: AlignmentIdentityBySampling()
440 * Date: SRE, Mon Oct 19 14:29:01 1998 [St. Louis]
442 * Purpose: Estimate and return the average pairwise
443 * fractional identity of an alignment,
446 * For use when there's so many sequences that
447 * an all vs. all rigorous calculation will
452 * Args: aseq - aligned sequences
453 * L - length of alignment
454 * N - number of seqs in alignment
455 * nsample - number of samples
457 * Returns: average fractional identity, 0..1.
460 AlignmentIdentityBySampling(char **aseq, int L, int N, int nsample)
462 int x, i, j; /* counters */
465 if (N < 2) return 1.0;
468 for (x = 0; x < nsample; x++)
471 do { j = CHOOSE(N); } while (j == i); /* make sure j != i */
472 sum += PairwiseIdentity(aseq[i], aseq[j]);
474 return sum / (float) nsample;
477 /* Function: MajorityRuleConsensus()
478 * Date: SRE, Tue Mar 7 15:30:30 2000 [St. Louis]
480 * Purpose: Given a set of aligned sequences, produce a
481 * majority rule consensus sequence. If >50% nonalphabetic
482 * (usually meaning gaps) in the column, ignore the column.
484 * Args: aseq - aligned sequences, [0..nseq-1][0..alen-1]
485 * nseq - number of sequences
486 * alen - length of alignment
488 * Returns: ptr to allocated consensus sequence.
489 * Caller is responsible for free'ing this.
492 MajorityRuleConsensus(char **aseq, int nseq, int alen)
494 char *cs; /* RETURN: consensus sequence */
495 int count[27]; /* counts for a..z and gaps in a column */
496 int idx,apos; /* counters for seq, column */
497 int spos; /* position in cs */
498 int x; /* counter for characters */
502 cs = MallocOrDie(sizeof(char) * (alen+1));
504 for (spos=0,apos=0; apos < alen; apos++)
506 for (x = 0; x < 27; x++) count[x] = 0;
508 for (idx = 0; idx < nseq; idx++)
510 if (isalpha(aseq[idx][apos])) {
511 sym = toupper(aseq[idx][apos]);
518 if ((float) count[26] / (float) nseq <= 0.5) {
520 for (x = 0; x < 26; x++)
521 if (count[x] > max) { max = count[x]; bestx = x; }
522 cs[spos++] = (char) ('A' + bestx);