void next_contig(GapIO *io, contig_t **cp) {
cache_decr(io, *cp);
if (++io->contig_num == io->db->Ncontigs)
io->contig_num = 0;
gio_read_contig(io, io->contig_num, cp);
cache_incr(io, *cp);
}
/*
* Exports Scaffold information to an AGP file
*
* Returns 0 on success
* -1 on failure
*/
int scaffold_to_agp(GapIO *io, char *fn) {
FILE *fp;
int i, j;
if (NULL == (fp = fopen(fn, "w+"))) {
verror(ERR_WARN, "scaffold_from_agp", "%s: %s", fn, strerror(errno));
return -1;
}
for (i = 0; io->scaffold && i < ArrayMax(io->scaffold); i++) {
scaffold_t *f = cache_search(io, GT_Scaffold,
arr(tg_rec, io->scaffold, i));
int start = 1, end = 1;
int k = 1;
if (!f) {
verror(ERR_WARN, "scaffold_from_agp", "Failed to load scaffold\n");
fclose(fp);
return -1;
}
cache_incr(io, f);
for (j = 0; f->contig && j < ArrayMax(f->contig); j++) {
scaffold_member_t *m = arrp(scaffold_member_t, f->contig, j);
contig_t *c = cache_search(io, GT_Contig, m->rec);
int ustart, uend;
int len;
/* Get the unpadded clipped contig length */
consensus_valid_range(io, m->rec, &ustart, &uend);
consensus_unpadded_pos(io, m->rec, uend, &uend);
len = uend - ustart + 1;
if (j) {
int gap = m->gap_size;
fprintf(fp, "%s\t%d\t%d\t%d\tN\t%d\tfragment\tyes\n",
f->name, start, start+gap-1, k++, gap);
start += gap;
}
fprintf(fp, "%s\t%d\t%d\t%d\tW\t%s\t%d\t%d\t+\n",
f->name, start, start + len-1,
k++, c->name, ustart, uend);
start += len;
}
cache_decr(io, f);
}
if (0 != fclose(fp)) {
verror(ERR_WARN, "scaffold_from_agp", "%s: %s", fn, strerror(errno));
return -1;
}
return 0;
}
void prev_contig(GapIO *io, contig_t **cp) {
cache_decr(io, *cp);
if (io->contig_num == 0)
io->contig_num = io->db->Ncontigs;
io->contig_num--;
gio_read_contig(io, io->contig_num, cp);
cache_incr(io, *cp);
}
static void tag_shift_for_delete(GapIO *io, tg_rec crec, tg_rec srec,
int start, int end, int pos, tg_rec brec) {
contig_iterator *ci;
rangec_t *r;
contig_t *c = cache_search(io, GT_Contig, crec);;
//printf("< tag in seq %"PRIrec" at %d\n", srec, pos);
cache_incr(io, c);
ci = contig_iter_new_by_type(io, crec, 0, CITER_FIRST | CITER_ISTART,
start+pos, end, GRANGE_FLAG_ISANNO);
if (!ci) {
cache_decr(io, c);
return;
}
while ((r = contig_iter_next(io, ci))) {
range_t r2, *r_out;
anno_ele_t *a;
bin_index_t *bin;
if (r->pair_rec != srec)
continue;
bin_remove_item(io, &c, GT_AnnoEle, r->rec);
r2.start = (r->start > start+pos) ? r->start-1 : r->start;
r2.end = r->end-1;
r2.mqual = r->mqual;
r2.rec = r->rec;
r2.pair_rec = r->pair_rec;
r2.flags = r->flags;
if (r2.end < r2.start) {
/* Tag entirely removed now, it must have been on a pad */
a = cache_search(io, GT_AnnoEle, r->rec);
a = cache_rw(io, a);
cache_deallocate(io, a);
continue;
}
bin = bin_add_to_range(io, &c, brec, &r2, &r_out, NULL, 0);
a = cache_search(io, GT_AnnoEle, r->rec);
if (a->bin != bin->rec /*||
a->idx != r_out - ArrayBase(range_t, bin->rng)*/) {
/* Annotation moved bins */
a = cache_rw(io, a);
a->bin = bin->rec;
//a->bin_idx = r_out - ArrayBase(range_t, bin->rng);
}
}
cache_decr(io, c);
contig_iter_del(ci);
}
/*
* Creates an anno_ele as per anno_ele_new, but also adds it to an object
* and creates the bin Range entry too.
*/
tg_rec anno_ele_add(GapIO *io, int obj_type, tg_rec obj_rec, tg_rec anno_rec,
int type, char *comment, int start, int end, char dir) {
range_t r;
anno_ele_t *e;
contig_t *c;
tg_rec crec;
bin_index_t *bin;
tg_rec seq_bin = 0;
/* Find contig for obj_rec/obj_type */
if (obj_type == GT_Contig) {
crec = obj_rec;
} else {
int st, en;
sequence_get_position2(io, obj_rec, &crec, &st, &en, NULL,
&seq_bin, NULL, NULL);
start += st;
end += st;
}
c = (contig_t *)cache_search(io, GT_Contig, crec);
cache_incr(io, c);
r.start = start;
r.end = end;
r.flags = GRANGE_FLAG_ISANNO;
r.mqual = type;
r.pair_rec = obj_rec;
if (GT_Seq == obj_type)
r.flags |= GRANGE_FLAG_TAG_SEQ;
r.rec = anno_ele_new(io, 0, obj_type, obj_rec, 0, type, dir, comment);
e = (anno_ele_t *)cache_search(io, GT_AnnoEle, r.rec);
e = cache_rw(io, e);
if (seq_bin)
bin = bin_add_to_range(io, &c, seq_bin, &r, NULL, NULL, 0);
else
bin = bin_add_range(io, &c, &r, NULL, NULL, 0);
if (!bin)
verror(ERR_FATAL, "anno_ele_add", "bin_add_to_range returned NULL");
e->bin = bin ? bin->rec : 0;
cache_decr(io, c);
return r.rec;
}
/*
* Tidies up after break contig or disassemble readings, looking for now
* redundant bins.
*
* This has the following functions (not all implemented yet!)
* 1) If a contig is totally empty, remove the contig.
* 2) If a bin is empty and all below it, remove the bin.
* 3) If a bin is empty and all above it, remove parent bins and link
* contig to new root. (TODO)
*/
static void remove_empty_bins(GapIO *io, tg_rec contig) {
contig_t *c = cache_search(io, GT_Contig, contig);
tg_rec first = 0;
cache_incr(io, c);
if (c->bin) {
if (remove_empty_bins_r(io, c->bin, &first)) {
cache_decr(io, c);
contig_destroy(io, contig);
return;
}
if (first != c->bin) {
bin_index_t *bin;
tg_rec bp, br, cdummy;
int offset;
/* Cut out the offending waste */
bin = cache_search(io, GT_Bin, first);
bin = cache_rw(io, bin);
bp = bin->parent;
// Find new bin offset
bin_get_position(io, bin, &cdummy, &offset);
assert(cdummy == contig);
bin->pos = offset;
bin->parent = contig;
bin->parent_type = GT_Contig;
bin->flags |= BIN_BIN_UPDATED;
c = cache_rw(io, c);
br = c->bin;
c->bin = first;
bin = cache_search(io, GT_Bin, bp);
bin = cache_rw(io, bin);
if (bin->child[0] == first) bin->child[0] = 0;
if (bin->child[1] == first) bin->child[1] = 0;
/* Recursively remove the bin tree from old root, br */
bin_destroy_recurse(io, br);
}
}
cache_decr(io, c);
}
/*
* Removes all tags of specific types (hashed in h, or all if h == NULL)
* from a specified contig.
*
* Returns 0 on success
* -1 on failure
*/
static int delete_tag_single_contig(GapIO *io, tg_rec crec,
HashTable *h, int verbose) {
contig_iterator *ci;
rangec_t *r;
contig_t *c;
int ret = -1;
ci = contig_iter_new_by_type(io, crec, 1, CITER_FIRST,
CITER_CSTART, CITER_CEND,
GRANGE_FLAG_ISANNO);
if (!ci)
return -1;
if (!(c = cache_search(io, GT_Contig, crec))) {
contig_iter_del(ci);
return -1;
}
cache_incr(io, c);
while (NULL != (r = contig_iter_next(io, ci))) {
char t[5];
(void)type2str(r->mqual, t);
if (!h || HashTableSearch(h, t, 4)) {
anno_ele_t *e;
if (verbose)
vmessage("Removing anno %s #%"PRIrec"\tContig %s\t%d..%d\n",
t, r->rec, c->name, r->start, r->end);
if (bin_remove_item(io, &c, GT_AnnoEle, r->rec)) goto fail;
/* FIXME: Need to reclaim the GT_AnnoEle record itself */
}
}
ret = 0;
fail:
contig_iter_del(ci);
cache_decr(io, c);
return ret;
}
/*
* ----------------------------------------------------------------------
* Remove Pad Columns. Sometimes we don't want to realign data, we just
* want to remove (aligned) columns of pads.
* ----------------------------------------------------------------------
*/
int remove_pad_columns(GapIO *io, int ncontigs, contig_list_t *contigs,
int percent_pad, int quiet) {
int i;
consensus_t *cons = NULL;
size_t max_alloc = 0;
for (i = 0; i < ncontigs; i++) {
tg_rec cnum = contigs[i].contig;
size_t len, j;
int ndel = 0;
contig_t *c;
if (!quiet) {
vmessage("Processing contig %d of %d (#%"PRIrec")\n",
i+1, ncontigs, cnum);
UpdateTextOutput();
}
c = cache_search(io, GT_Contig, cnum);
if (!c)
return -1;
cache_incr(io, c);
len = contigs[i].end - contigs[i].start + 1;
if (max_alloc < len) {
max_alloc = len;
cons = realloc(cons, max_alloc * sizeof(*cons));
}
if (0 != calculate_consensus(io, cnum,
contigs[i].start, contigs[i].end,
cons)) {
free(cons);
cache_decr(io, c);
return -1;
}
for (j = 0; j < len; j++) {
if (cons[j].call != 4)
continue;
if (100 * cons[j].counts[4] / cons[j].depth < percent_pad)
continue;
if (!quiet)
vmessage(" Removing column %d %d%% pad (%d of %d), conf. %f)\n",
(int)j+contigs[i].start,
100 * cons[j].counts[4] / cons[j].depth,
cons[j].counts[4], cons[j].depth,
cons[j].scores[cons[j].call]);
contig_delete_base(io, &c, contigs[i].start + j - ndel);
ndel++;
}
cache_decr(io, c);
}
if (cons)
free(cons);
return 0;
}
/*
* Takes a multiple alignment and updates the on-disk data structures to
* match. This needs to correct confidence values, original positions and
* tags too.
*/
void update_io(GapIO *io, tg_rec cnum, MALIGN *malign, Array indels) {
CONTIGL *cl;
tg_rec rnum;
range_t r, *r_out;
bin_index_t *bin;
contig_t *c = cache_search(io, GT_Contig, cnum);
size_t i, nindel;
cache_incr(io, c);
/*
* To minimise number of data modifications we use a three step approach.
*
* Step 1: insert columns of pads, shifting reads as appropriate.
* Step 2: edit sequence alignments as required, possibly involving
* moving sequences and/or adding and removing pads.
* Step 3: remove columns of entire pads.
*
* This means that when we introduce a column of pads we don't have
* to make edits to every single read position down stream, and can
* instead make use of the optimised recursive bin functions to do this
* for us.
*/
/* Step 1: make indels */
nindel = ArrayMax(indels);
for (i = 0; i < nindel; i++) {
con_indel_t *id = arrp(con_indel_t, indels, i);
int j;
if (id->size > 0) {
contig_insert_bases(io, &c, id->pos+1, '*', -1, id->size);
} else {
for (j = 0; j < -id->size; j++) {
contig_delete_pad(io, &c, id->pos+1);
}
}
}
/* Step 2: edit alignments */
for (cl = malign->contigl; cl; cl = cl->next) {
seq_t *s, *sorig;
int len, update_range = 0;
int shift;
rnum = cl->id;
sorig = cache_search(io, GT_Seq, rnum);
cache_incr(io, sorig);
s = dup_seq(sorig);
if (cl->mseg->comp)
complement_seq_t(s);
len = s->right - s->left + 1;
/* Check if sequence has changed. If so assign a new one */
if (cl->mseg->length != len ||
memcmp(s->seq + s->left-1, cl->mseg->seq, cl->mseg->length) != 0) {
int newlen = s->left-1 + ABS(s->len) - s->right + cl->mseg->length;
int i, j, np;
char *newseq = malloc(newlen+1);
int8_t *newconf = malloc(newlen+1);
/* Build new seq/conf arrays */
memcpy(newseq, s->seq, s->left-1);
memcpy(newconf, s->conf, s->left-1);
memcpy(&newseq[s->left-1], cl->mseg->seq, cl->mseg->length);
/*
* Step through both old and new sequences working out how
* they differ. This will (*should*) be entire pad movements.
* i = index to old seq
* j = index to new seq
* np = number of pads added minus removed from old seq.
*/
np = 0;
for (i =j =s->left-1;
i < ABS(s->len) && j < s->left-1 + cl->mseg->length;
) {
/* Bases match */
if (toupper(newseq[j]) == toupper(s->seq[i]) ||
(s->seq[i] == '.' && newseq[j] == 'N')) {
if (isupper(s->seq[i]))
newseq[j] = toupper(newseq[j]);
else
newseq[j] = tolower(newseq[j]);
newconf[j] = s->conf[i];
i++, j++;
continue;
}
/* Pad removed */
if (s->seq[i] == '*') {
i++;
tag_shift_for_delete(io, cnum, rnum, cl->mseg->offset,
cl->mseg->length, i+np--,
s->bin);
/*
if (io_length(io, rnum) < 0) {
tag_shift_for_delete(io, rnum, r.length - i + 1);
} else {
tag_shift_for_delete(io, rnum, i+np--);
}
*/
continue;
}
/* Pad created */
if (newseq[j] == '*') {
int k;
int ql = 0, qr = 0;
for (k = i-1; k >= 0; k--) {
if (s->seq[k] != '*') {
ql = s->conf[k];
break;
}
}
for (k = i+1; k < s->right; k++) {
if (s->seq[k] != '*') {
qr = s->conf[k];
break;
}
}
newconf[j] = MIN(ql, qr); /* min conf of neighbours */
j++;
tag_shift_for_insert(io, cnum, rnum, cl->mseg->offset,
cl->mseg->length, i+ ++np,
s->bin);
/*
if (io_length(io, rnum) < 0) {
tag_shift_for_insert(io, rnum, r.length - i + 1);
} else {
tag_shift_for_insert(io, rnum, i+ ++np);
}
*/
continue;
}
fprintf(stderr, "Alignment introduced non-pad character");
abort();
}
/* Pads previously at the end of the reading & now removed */
while (i < s->right) {
if (s->seq[i] == '*') {
i++;
tag_shift_for_delete(io, cnum, rnum, cl->mseg->offset,
cl->mseg->length, i+np--,
s->bin);
/*
if (io_length(io, rnum) < 0) {
tag_shift_for_delete(io, rnum, r.length - i + 1);
} else {
tag_shift_for_delete(io, rnum, i+np--);
}
*/
} else {
/* Error: clipped data that wasn't a pad */
abort();
}
}
/* Should only be pads remaining in newseq, if anything */
s->right = j;
for (; j < s->left-1 + cl->mseg->length; j++) {
if (newseq[j] != '*') {
fprintf(stderr, "Alignment introduced non-pad character");
abort();
}
newconf[j] = 0;
}
/* Append on the right hand cutoff data */
for (; i < ABS(s->len); i++, j++) {
newseq[j] = s->seq[i];
newconf[j] = s->conf[i];
}
if (j != newlen) {
abort();
}
/* Write it back out */
/* Copy newseq/newconf into seq_t */
s->seq = newseq;
s->conf = newconf;
update_range = 0;
if (ABS(s->len) != j) {
/* Length change implies updating the range array too */
s->len = s->len >= 0 ? j : -j;
update_range = 1;
}
if (cl->mseg->comp)
complement_seq_t(s);
/* The memcpy trashes the block pointer, so special care needed */
{
sorig = cache_rw(io, sorig);
void *blk = sorig->block;
memcpy(sorig, s, sizeof(seq_t));
sorig->block = blk;
}
if (update_range)
sorig = cache_item_resize(sorig, sizeof(*sorig) +
sequence_extra_len(sorig));
sequence_reset_ptr(sorig);
if (s->name)
memcpy(sorig->name, s->name, s->name_len+1);
if (s->trace_name)
memcpy(sorig->trace_name, s->trace_name, s->trace_name_len+1);
if (s->alignment)
memcpy(sorig->alignment, s->alignment, s->alignment_len+1);
memcpy(sorig->seq, s->seq, ABS(s->len));
memcpy(sorig->conf, s->conf, ABS(s->len));
xfree(newconf);
xfree(newseq);
}
{
int st, en, or;
sequence_get_position(io, s->rec, NULL, &st, &en, &or);
if (or ^ (sorig->len < 0)) {
shift = ABS(sorig->len) - sorig->right;
} else {
shift = sorig->left-1;
}
st += shift;
if (st != cl->mseg->offset+1) {
update_range = 1;
}
}
free(s);
if (update_range) {
int bin_changed = 0;
/* Get old range and pair data */
s = sorig;
bin = cache_search(io, GT_Bin, s->bin);
r = *arrp(range_t, bin->rng, s->bin_index);
assert(r.rec == s->rec);
/* Update range, tedious and slow way */
bin_remove_item(io, &c, GT_Seq, s->rec);
r.start = cl->mseg->offset + 1 - shift;
r.end = r.start + ABS(s->len) - 1;
bin = bin_add_range(io, &c, &r, &r_out, NULL, 0);
/* Check if the new bin has a different complemented status too */
if (s->bin != bin->rec) {
int old_comp = bin_get_orient(io, s->bin);
int new_comp = bin_get_orient(io, bin->rec);
if (new_comp != old_comp) {
//int tmp;
s = cache_rw(io, s);
s->len *= -1;
s->flags ^= SEQ_COMPLEMENTED;
//tmp = s->left;
//s->left = ABS(s->len) - (s->right-1);
//s->right = ABS(s->len) - (tmp-1);
}
bin_changed = 1;
}
/* Update seq bin & bin_index fields */
s = cache_rw(io, s);
s->bin = bin->rec;
s->bin_index = r_out - ArrayBase(range_t, bin->rng);
if (bin_changed) {
if (-1 == sequence_fix_anno_bins(io, &s)) {
verror(ERR_WARN, "update_io",
"sequence_fix_anno_bins() failure");
}
}
}
cache_decr(io, sorig);
}
/* Step 3 (remove pad columns) done in calling function. */
cache_decr(io, c);
}
int main(int argc, char **argv) {
GapIO *io;
int xpos = 0;
int opt;
int lp_mode = 0;
int mode = DISPLAY_QUAL | DISPLAY_CUTOFFS;
extern char *optarg;
contig_t *c;
int cnum = 0;
int read_only = 1;
while ((opt = getopt(argc, argv, "hl:dcCx:e")) != -1) {
switch (opt) {
case '?':
case 'h':
usage();
return 0;
case 'd':
mode |= DISPLAY_DIFFS;
break;
case 'c':
mode |= DISPLAY_CUTOFFS;
break;
case 'C':
mode &= ~DISPLAY_CUTOFFS;
break;
case 'l':
lp_mode = atoi(optarg);
break;
case 'x':
cnum = atoi(optarg)-1;
break;
case 'e':
read_only = 0;
break;
default:
if (opt == ':')
fprintf(stderr, "Missing parameter\n");
else
fprintf(stderr, "Unknown option '%c'\n", opt);
usage();
return 1;
}
}
if (optind == argc) {
usage();
return 1;
}
if (NULL == (io = gio_open(argv[optind], read_only, 0))) {
fprintf(stderr, "Unable to open db: %s\n", argv[1]);
return 1;
}
optind++;
if (optind != argc) {
xpos = atoi(argv[optind]);
}
io->contig_num = cnum;
gio_read_contig(io, cnum, &c);
cache_incr(io, c);
#ifdef TEST_MODE
//test_mode(io, &c, xpos);
test_mode2(io, &c, xpos);
//benchmark(io, &c);
//test_mode3(io, arr(GCardinal, io->contig_order, cnum), xpos);
#endif
if (lp_mode) {
print_output(io, &c, xpos, lp_mode, mode);
gio_close(io);
} else {
init_curses();
curses_loop(io, &c, xpos, mode);
endwin();
if (io->cache && io->debug_level > 0) {
fputs("\n=== cache ===", stderr);
HacheTableStats(io->cache, stderr);
}
gio_close(io);
}
if (!lp_mode) {
printf("\n\n\tg_view:\tShort Read Alignment Viewer, version 1.2.11"SVN_VERS"\n");
printf("\n\tAuthor:\tJames Bonfield ([email protected])\n");
printf("\t\t2007-2011, Wellcome Trust Sanger Institute\n\n");
}
return 0;
}
int edview_search_cons_discrep(edview *xx, int dir, int strand, char *value) {
int start, end;
int found = 0, at_end = 0;
int fpos, i;
double qval = atof(value);
consensus_t cons[WIN_WIDTH+1];
contig_t *c;
/* Set initial start positions */
if (dir) {
start = xx->cursor_apos + (dir ? 1 : -1);
end = start + (WIN_WIDTH-1);
} else {
end = xx->cursor_apos + (dir ? 1 : -1);
start = end - (WIN_WIDTH-1);
}
fpos = xx->cursor_apos;
/* Loop WIN_WIDTH block at a time */
c = cache_search(xx->io, GT_Contig, xx->cnum);
cache_incr(xx->io, c);
do {
calculate_consensus(xx->io, xx->cnum, start, end, cons);
if (dir) {
for (i = 0; i < WIN_WIDTH; i++) {
if (cons[i].discrep >= qval) {
found = 1;
break;
}
}
} else {
for (i = WIN_WIDTH-1; i; i--) {
if (cons[i].discrep >= qval) {
found = 1;
break;
}
}
}
if (found) {
fpos = start + i;
break;
}
/* Next search region - overlapping by patlen+pads */
if (dir) {
start += WIN_WIDTH;
end += WIN_WIDTH;
if (start > c->end)
at_end = 1;
} else {
start -= WIN_WIDTH;
end -= WIN_WIDTH;
if (end < c->start)
at_end = 1;
}
} while (!at_end);
cache_decr(xx->io, c);
if (found) {
edSetCursorPos(xx, GT_Contig, xx->cnum, fpos, 1);
return 0;
}
return -1;
}
/*
* A recursive break contig function.
* bin_num The current bin being moved or split.
* pos The contig break point.
* offset The absolute positional offset of this bin in original contig
* pleft The parent bin/contig record num in the left new contig
* pright The parent bin/contig record num in the right new contig
* child_no 0 or 1 - whether this bin is the left/right child of its parent
*/
static int break_contig_recurse(GapIO *io, HacheTable *h,
contig_t *cl, contig_t *cr,
tg_rec bin_num, int pos, int offset,
int level, tg_rec pleft, tg_rec pright,
int child_no, int complement) {
int i, j, f_a, f_b;
tg_rec rbin;
bin_index_t *bin = get_bin(io, bin_num), *bin_dup ;
//int bin_min, bin_max;
int nseqs;
tg_rec opright; /* old pright, needed if we revert back */
cache_incr(io, bin);
if (bin->flags & BIN_COMPLEMENTED) {
complement ^= 1;
}
if (complement) {
f_a = -1;
f_b = offset + bin->size-1;
} else {
f_a = +1;
f_b = offset;
}
printf("%*sBreak offset %d pos %d => test bin %"PRIrec": %d..%d\n",
level*4, "",
offset, pos, bin->rec,
NMIN(bin->start_used, bin->end_used),
NMAX(bin->start_used, bin->end_used));
bin = cache_rw(io, bin);
nseqs = bin->nseqs;
bin->nseqs = 0;
/* Invalidate any cached data */
bin_invalidate_track(io, bin, TRACK_ALL);
if (bin->flags & BIN_CONS_VALID) {
bin->flags |= BIN_BIN_UPDATED;
bin->flags &= ~BIN_CONS_VALID;
}
//bin_min = bin->rng ? NMIN(bin->start_used, bin->end_used) : offset;
//bin_max = bin->rng ? NMAX(bin->start_used, bin->end_used) : offset;
/*
* Add to right parent if this bin is to the right of pos,
* or if the used portion is to the right and we have no left child.
*
* FIXME: Not a valid assumption!
* The used portion of a bin is not a placeholder for the used portion
* of all the the children beneath it. Therefore if the used portion of
* this bin is > pos (and we have no left child) it still doesn't mean
* that the absolute positions of the used portion of the right child
* won't be < pos.
*/
if (offset >= pos /*|| (bin_min >= pos && !bin->child[0])*/) {
printf("%*sADD_TO_RIGHT pl=%"PRIrec" pr=%"PRIrec"\n",
level*4, "", pleft, pright);
if (0 != break_contig_move_bin(io, bin,
cl, pleft, cr, pright,
child_no))
return -1;
bin_incr_nseq(io, bin, nseqs);
cache_decr(io, bin);
return 0;
}
/*
* Add to left parent if this bin is entirely to the left of pos,
* or if the used portion is to the left and we have no right child.
*/
if (offset + bin->size < pos /*|| (bin_max < pos && !bin->child[1])*/) {
printf("%*sADD_TO_LEFT\n", level*4, "");
//if (0 != break_contig_move_bin(io, bin, cr, pright, cl, pleft, child_no))
//return -1;
bin_incr_nseq(io, bin, nseqs);
cache_decr(io, bin);
return 0;
}
/*
* Nominally the bin overlaps both left and right and so needs duplicating.
* There are cases though at the roots of our trees where duplicating is
* unnecessary as it leads to empty bins at the root. In this case
* we skip creating a duplicate for the right, or alternatively steal
* the left root bin and use that instead.
*
* Similarly the range_t array will either be left where it is, moved to
* the right contig, or split in half (creating a new one for the right).
*
* FIXED: always need this. Eg:
*
* |-------------empty--------------|
* |----------------|---------------|
* |--------|-------|--------|------|
* ^
* |
* break here
*
* In this case we need to duplicate the parent as it overlaps the left
* bin, which may (or may not) have data that needs to end up in the right
* hand contig. Just duplicate for now and free later on if needed.
*/
if (1 /* always! */ || pright != cr->rec ||
(bin->rng && NMAX(bin->start_used, bin->end_used) >= pos)) {
//printf("NMAX=%d >= %d\n", NMAX(bin->start_used, bin->end_used), pos);
rbin = 0;
/* Possibly steal left contig's bin */
if (pleft == cl->rec && NMIN(bin->start_used, bin->end_used) >= pos) {
#if 0
/* Currently this doesn't always work */
if (bin->child[1]) {
bin_index_t *ch = get_bin(io, bin->child[1]);
if (NMIN(ch->pos, ch->pos + ch->size-1) >= pos) {
rbin = cl->bin;
cl->bin = bin->child[0];
}
}
#else
pleft = bin->rec;
#endif
} else {
pleft = bin->rec;
}
/* Create new bin, or use root of contig if it's unused so far */
if (!rbin && pright == cr->rec) {
rbin = cr->bin;
}
/* Otherwise we genuingly need a duplicate */
if (!rbin)
rbin = bin_new(io, 0, 0, 0, GT_Bin);
/* Initialise with duplicate values from left bin */
bin_dup = get_bin(io, rbin);
bin_dup = cache_rw(io, bin_dup);
bin_dup->size = bin->size;
bin_dup->pos = bin->pos;
bin_dup->parent = pright;
bin_dup->parent_type = (pright == cr->rec ? GT_Contig : GT_Bin);
bin_dup->flags = bin->flags | BIN_BIN_UPDATED;
bin_dup->start_used = bin->start_used;
bin_dup->end_used = bin->end_used;
/*
* Shift bin to offset if it's the contig root.
* It'll be shifted back by the correct amount later.
*/
if (pright == cr->rec) {
printf("moving root bin to offset=%d comp=%d\n", offset, complement);
bin_dup->pos = offset;
}
printf("%*sCreated dup for right, rec %"PRIrec"\n",
level*4,"", bin_dup->rec);
break_contig_move_bin(io, bin_dup, cl, 0, cr, pright, child_no);
opright = pright;
pright = bin_dup->rec;
} else {
bin_dup = NULL;
pleft = bin->rec;
}
if (!bin->rng) {
/* Empty bin */
printf("%*sEMPTY range\n", level*4, "");
bin->start_used = bin->end_used = 0;
bin->flags |= BIN_BIN_UPDATED;
if (bin_dup) {
bin_dup->start_used = bin_dup->end_used = 0;
bin_dup->flags |= BIN_BIN_UPDATED;
}
} else if (NMIN(bin->start_used, bin->end_used) >= pos) {
/* Move range to right contig */
printf("%*sDUP %"PRIrec", MOVE Array to right\n",
level*4, "", bin_dup->rec);
bin_dup->rng = bin->rng;
bin_dup->rng_rec = bin->rng_rec;
bin_dup->rng_free = bin->rng_free;
if (bin_dup->rng_rec)
bin_dup->flags |= BIN_RANGE_UPDATED;
if (bin->rec != bin_dup->rec) {
bin->rng = NULL;
bin->rng_rec = 0;
bin->rng_free = -1;
bin->flags |= BIN_BIN_UPDATED;
}
bin->start_used = bin->end_used = 0;
break_contig_reparent_seqs(io, bin_dup);
if (bin_dup->rng) {
int n = ArrayMax(bin_dup->rng);
for (i = j = 0; i < n; i++) {
range_t *r = arrp(range_t, bin_dup->rng, i), *r2;
if (r->flags & GRANGE_FLAG_UNUSED)
continue;
if ((r->flags & GRANGE_FLAG_ISMASK) != GRANGE_FLAG_ISANNO) {
HacheData hd; hd.i = 1;
HacheTableAdd(h, (char *)&r->rec, sizeof(r->rec), hd,NULL);
j++;
}
}
bin_incr_nseq(io, bin_dup, j);
}
} else if (NMAX(bin->start_used, bin->end_used) < pos) {
/* Range array already in left contig, so do nothing */
printf("%*sMOVE Array to left\n", level*4, "");
if (bin_dup)
bin_dup->start_used = bin_dup->end_used = 0;
if (bin->rng) {
int n = ArrayMax(bin->rng);
for (i = j = 0; i < n; i++) {
range_t *r = arrp(range_t, bin->rng, i);
if (r->flags & GRANGE_FLAG_UNUSED)
continue;
if ((r->flags & GRANGE_FLAG_ISMASK) != GRANGE_FLAG_ISANNO) {
HacheData hd; hd.i = 0;
HacheTableAdd(h, (char *)&r->rec, sizeof(r->rec), hd,NULL);
j++;
}
}
bin_incr_nseq(io, bin, j);
}
} else {
/* Range array covers pos, so split in two */
int n, nl = 0, nr = 0;
int lmin = bin->size, lmax = 0, rmin = bin->size, rmax = 0;
printf("%*sDUP %"PRIrec", SPLIT array\n", level*4, "", bin_dup->rec);
bin->flags |= BIN_RANGE_UPDATED;
bin_dup->flags |= BIN_RANGE_UPDATED;
bin_dup->rng = ArrayCreate(sizeof(range_t), 0);
bin_dup->rng_free = -1;
/* Pass 1 - hash sequences */
n = ArrayMax(bin->rng);
for (i = 0; i < n; i++) {
range_t *r = arrp(range_t, bin->rng, i);
int cstart; /* clipped sequence positions */
seq_t *s;
if (r->flags & GRANGE_FLAG_UNUSED)
continue;
if ((r->flags & GRANGE_FLAG_ISMASK) == GRANGE_FLAG_ISANNO)
continue;
s = (seq_t *)cache_search(io, GT_Seq, r->rec);
if ((s->len < 0) ^ complement) {
cstart = NMAX(r->start, r->end) - (s->right-1);
} else {
cstart = NMIN(r->start, r->end) + s->left-1;
}
if (cstart >= pos) {
HacheData hd; hd.i = 1;
HacheTableAdd(h, (char *)&r->rec, sizeof(r->rec), hd, NULL);
} else {
HacheData hd; hd.i = 0;
HacheTableAdd(h, (char *)&r->rec, sizeof(r->rec), hd, NULL);
}
}
/* Pass 2 - do the moving of anno/seqs */
n = ArrayMax(bin->rng);
for (i = j = 0; i < n; i++) {
range_t *r = arrp(range_t, bin->rng, i), *r2;
int cstart; /* clipped sequence positions */
if (r->flags & GRANGE_FLAG_UNUSED)
continue;
if ((r->flags & GRANGE_FLAG_ISMASK) == GRANGE_FLAG_ISANNO) {
cstart = NMAX(r->start, r->end);
} else {
seq_t *s = (seq_t *)cache_search(io, GT_Seq, r->rec);
if ((s->len < 0) ^ complement) {
cstart = NMAX(r->start, r->end) - (s->right-1);
} else {
cstart = NMIN(r->start, r->end) + s->left-1;
}
}
if (cstart >= pos &&
((r->flags & GRANGE_FLAG_ISMASK) == GRANGE_FLAG_ISANNO)) {
anno_ele_t *a = (anno_ele_t *)cache_search(io,
GT_AnnoEle,
r->rec);
/* If it's an annotation on a sequence < pos then we
* still don't move.
*
* FIXME: we have no guarantee that the sequence being
* annotated is in the same bin as this annotation, as
* they may be different sizes and end up in different
* bins. (Should we enforce anno always in same bin as seq?
* If so, consensus annos fit anywhere?)
*/
if (a->obj_type == GT_Seq) {
HacheItem *hi = HacheTableSearch(h,
(char *)&r->pair_rec,
sizeof(r->pair_rec));
if (hi) {
if (hi->data.i == 0)
cstart = pos-1;
} else {
puts("FIXME: annotation for seq in unknown place - "
"work out correct location and move if needed.");
}
}
}
if (cstart >= pos) {
r2 = (range_t *)ArrayRef(bin_dup->rng, ArrayMax(bin_dup->rng));
*r2 = *r;
if (rmin > r->start) rmin = r->start;
if (rmin > r->end) rmin = r->end;
if (rmax < r->start) rmax = r->start;
if (rmax < r->end) rmax = r->end;
if ((r->flags & GRANGE_FLAG_ISMASK) == GRANGE_FLAG_ISSEQ)
nr++;
} else {
if (lmin > r->start) lmin = r->start;
if (lmin > r->end) lmin = r->end;
if (lmax < r->start) lmax = r->start;
if (lmax < r->end) lmax = r->end;
if (j != i) {
r2 = arrp(range_t, bin->rng, j);
*r2 = *r;
}
j++;
if ((r->flags & GRANGE_FLAG_ISMASK) == GRANGE_FLAG_ISSEQ)
nl++;
}
}
bin_incr_nseq(io, bin, nl);
bin_incr_nseq(io, bin_dup, nr);
ArrayMax(bin->rng) = j;
#if 0
/*
* Right now this causes problems, but I'm not sure why. Try again
* after we've fixed the bin->nseqs issues and other deallocation
* woes.
*/
if (ArrayMax(bin_dup->rng) == 0 && bin_dup->parent_type == GT_Bin) {
/* We didn't need it afterall! Odd. */
bin_index_t *pb;
printf("Purging bin %d that we didn't need afterall\n",
bin_dup->rec);
cache_rec_deallocate(io, GT_Bin, bin_dup->rec);
pb = cache_search(io, GT_Bin, bin_dup->parent);
if (pb->child[0] == bin_dup->rec)
pb->child[0] = 0;
if (pb->child[1] == bin_dup->rec)
pb->child[1] = 0;
bin_dup = NULL;
pright = opright;
}
#endif
if (bin_dup)
break_contig_reparent_seqs(io, bin_dup);
if (lmin < lmax) {
bin->start_used = lmin;
bin->end_used = lmax;
} else {
/* No data left in bin */
bin->start_used = 0;
bin->end_used = 0;
}
printf("%*sLeft=>%d..%d right=>%d..%d\n", level*4, "",
lmin, lmax, rmin, rmax);
if (bin_dup) {
if (rmin < rmax) {
bin_dup->start_used = rmin;
bin_dup->end_used = rmax;
} else {
/* No data moved in bin */
bin_dup->start_used = 0;
bin_dup->end_used = 0;
}
}
}
/* Recurse */
for (i = 0; i < 2; i++) {
bin_index_t *ch;
if (!bin->child[i])
continue;
ch = get_bin(io, bin->child[i]);
if (0 != break_contig_recurse(io, h, cl, cr, bin->child[i], pos,
NMIN(ch->pos, ch->pos + ch->size-1),
level+1, pleft, pright,
i, complement))
return -1;
}
cache_decr(io, bin);
// if (bin_dup)
// cache_decr(io, bin_dup);
return 0;
}
/*
* Open a database, optionally in read-only mode and creating if desired too.
*
* Returns GapIO pointer to DB on success.
* NULL on failure.
*/
GapIO *gio_open(char *fn, int ro, int create) {
GapIO *io = (GapIO *)calloc(1, sizeof(*io));
char *cp;
io->iface = get_iface_g();
if (create) {
if (-1 == io->iface->create(fn))
return NULL;
}
io->min_bin_size = MIN_BIN_SIZE; /* default */
/* Initialise the cache */
cache_create(io);
if (NULL == (io->dbh = io->iface->connect(fn, ro))) {
if (!ro) {
ro = 1;
if (NULL == (io->dbh = io->iface->connect(fn, ro)))
return NULL;
} else {
return NULL;
}
}
io->read_only = ro;
if (create) {
io->iface->database.create(io->dbh, NULL);
}
/* Cache the GDatabase struct */
if (NULL == (io->db = cache_search(io, GT_Database, 0)))
return NULL;
cache_incr(io, io->db);
if (io->db->version > DB_VERSION) {
verror(ERR_WARN, "Open Database",
"Database version %d is too new for this version of gap5",
io->db->version);
return NULL;
}
/* Load the contigs array */
io->contig_order = cache_search(io, GT_RecArray, io->db->contig_order);
cache_incr(io, io->contig_order);
/* Load the library array */
io->library = cache_search(io, GT_RecArray, io->db->library);
cache_incr(io, io->library);
/* Initialise the contig and cursor registration hashes */
contig_register_init(io);
io->iface->setopt(io->dbh, OPT_COMP_MODE, COMP_MODE_ZLIB);
/* Copy the name */
if (NULL == (cp = strrchr(fn, '/')))
cp = fn;
io->name = strdup(cp);
return io;
}
/*
* Complements a scaffold; both complementing each contig within it and
* reversing the order of contigs in the scaffold.
*
* Returns 0 on success
* -1 on failure
*/
int complement_scaffold(GapIO *io, tg_rec srec) {
scaffold_t *f;
int i, j, nc = ArrayMax(io->contig_order);
scaffold_member_t *contigs;
tg_rec *crecs;
HashTable *h;
reg_order ro;
reg_buffer_start rs;
reg_buffer_end re;
if (!(f = cache_search(io, GT_Scaffold, srec)))
return -1;
if (!(f = cache_rw(io, f)))
return -1;
cache_incr(io, f);
/* Complement contigs */
contigs = ArrayBase(scaffold_member_t, f->contig);
for (i = 0; i < ArrayMax(f->contig); i++) {
complement_contig(io, contigs[i].rec);
}
/* Reverse the order of the contigs in the scaffold array */
for (i = 0, j = ArrayMax(f->contig)-1; i < j; i++, j--) {
scaffold_member_t cr1 = contigs[i];
contigs[i] = contigs[j];
contigs[j] = cr1;
}
/*
* Reverse the order of contigs in the contig_order array too.
* This is the part that really matters. It's also hard as the contigs
* in the contig order array could be in any order and not adjacent.
* For our purposes we'll just ensure the contigs in this scaffold in
* the contig order array match our freshly complemented scaffold
* ordering.
*
* We initially build a hash table of contigs in this scaffold, and
* then iterate through contig_order copying out the new contigs whenever
* one matches.
*/
h = HashTableCreate(nc, 0);
for (i = 0; i < ArrayMax(f->contig); i++) {
HashData hd;
hd.i = 0;
HashTableAdd(h, (char *)&contigs[i].rec, sizeof(tg_rec), hd, NULL);
}
/* Replace any contig matching the scaffold with the new order */
crecs = ArrayBase(tg_rec, io->contig_order);
for (i = j = 0; i < nc; i++) {
HashItem *hi;
if (!(hi = HashTableSearch(h, (char *)&crecs[i], sizeof(tg_rec))))
continue;
crecs[i] = contigs[j++].rec;
}
/* Send event messages around */
rs.job = REG_BUFFER_START;
for (i = 0; i < nc; i++) {
HashItem *hi;
if (!(hi = HashTableSearch(h, (char *)&crecs[i], sizeof(tg_rec))))
continue;
contig_notify(io, crecs[i], (reg_data *)&rs);
}
ro.job = REG_ORDER;
for (i = 0; i < nc; i++) {
HashItem *hi;
if (!(hi = HashTableSearch(h, (char *)&crecs[i], sizeof(tg_rec))))
continue;
ro.pos = i+1;
contig_notify(io, crecs[i], (reg_data *)&ro);
}
/* Notify the end of our updates */
re.job = REG_BUFFER_END;
for (i = 0; i < nc; i++) {
HashItem *hi;
if (!(hi = HashTableSearch(h, (char *)&crecs[i], sizeof(tg_rec))))
continue;
contig_notify(io, crecs[i], (reg_data *)&re);
}
HashTableDestroy(h, 0);
cache_decr(io, f);
return 0;
}
static void parse_incr(struct req_info *req)
{
int cres, cache_only, rv;
const unsigned char *key;
uint32_t ksize;
int64_t increment, newval;
const int max = 65536;
/* Request format:
* 4 ksize
* ksize key
* 8 increment (big endian int64_t)
*/
ksize = * (uint32_t *) req->payload;
ksize = ntohl(ksize);
/* Sanity check on sizes:
* - ksize + 8 must be < req->psize
* - ksize + 8 must be < 2^16 = 64k
*/
if ( (req->psize < ksize + 8) || ((ksize + 8) > max)) {
stats.net_broken_req++;
req->reply_err(req, ERR_BROKEN);
return;
}
if (settings.read_only) {
req->reply_err(req, ERR_RO);
return;
}
FILL_CACHE_FLAG(incr);
key = req->payload + sizeof(uint32_t);
increment = ntohll( * (int64_t *) (key + ksize) );
cres = cache_incr(cache_table, key, ksize, increment, &newval);
if (cres == -3) {
req->reply_err(req, ERR_MEM);
return;
} else if (cres == -2) {
/* the value was not NULL terminated */
req->reply_mini(req, REP_NOMATCH);
return;
}
if (!cache_only) {
/* at this point, the cache_incr() was either successful or a
* miss, but we don't really care */
rv = put_in_queue(req, REQ_INCR, 1, key, ksize,
(unsigned char *) &increment,
sizeof(increment));
if (!rv) {
req->reply_err(req, ERR_MEM);
return;
}
} else {
if (cres == -1) {
req->reply_mini(req, REP_NOTIN);
} else {
newval = htonll(newval);
req->reply_long(req, REP_OK,
(unsigned char *) &newval,
sizeof(newval));
}
}
return;
}
/*
* Open a database, optionally in read-only mode and creating if desired too.
*
* Returns GapIO pointer to DB on success.
* NULL on failure.
*/
GapIO *gio_open(char *fn, int ro, int create) {
GapIO *io = (GapIO *)calloc(1, sizeof(*io));
char *cp;
int lock_err;
/* Check locks */
lock_err = actf_lock(ro, fn, create);
if (!create && (lock_err == 3 || lock_err == 5)) {
vmessage("Opening database in read only mode instead.\n");
ro = 1;
lock_err = actf_lock(ro, fn, create);
}
if (lock_err != 0) {
verror(ERR_WARN, "Open Database",
"Unable to lock and/or open the database.");
return NULL;
}
io->iface = get_iface_g();
if (create) {
if (0 != io->iface->create(fn)) {
xperror("In tg_gio.c:gio_open()", xperror_fatal);
return NULL;
}
}
io->min_bin_size = MIN_BIN_SIZE; /* default */
/* Initialise the cache */
cache_create(io);
if (NULL == (io->dbh = io->iface->connect(fn, ro))) {
if (!ro) {
ro = 1;
if (NULL == (io->dbh = io->iface->connect(fn, ro)))
return NULL;
} else {
return NULL;
}
}
io->read_only = ro;
if (create) {
io->iface->database.create(io->dbh, NULL, db_version);
}
/* Cache the GDatabase struct */
if (NULL == (io->db = cache_search(io, GT_Database, 0)))
return NULL;
cache_incr(io, io->db);
if (io->db->version > DB_VERSION) {
verror(ERR_WARN, "Open Database",
"Database version %d is too new for this version of gap5",
io->db->version);
return NULL;
}
/* Load the contigs array */
io->contig_order = cache_search(io, GT_RecArray, io->db->contig_order);
cache_incr(io, io->contig_order);
/* Load the scaffold array */
if (io->db->scaffold) {
io->scaffold =
cache_search(io, GT_RecArray, io->db->scaffold);
cache_incr(io, io->scaffold);
} else {
/* FIXME: create a dummy order of 1 per contig? */
io->scaffold = 0;
}
/* Load the library array */
io->library = cache_search(io, GT_RecArray, io->db->library);
cache_incr(io, io->library);
/* Initialise the contig and cursor registration hashes */
contig_register_init(io);
io->iface->setopt(io->dbh, OPT_COMP_MODE, COMP_MODE_ZLIB);
/* Copy the name */
if ((cp = strrchr(fn, '/')))
cp++;
else
cp = fn;
io->name = strdup(cp);
io->last_bin = 0;
io->incr_svalue = io->incr_rvalue = io->incr_avalue = 0;
io->max_template_size = 0;
io->debug_level = 0;
io->debug_fp = stderr;
//update_uniqueness_hash(io);
#ifdef DO_LOGGING
open_log_file(io, fn);
#endif
return io;
}
/*
* Breaks a contig in two such that snum is the right-most reading of
* a new contig.
*/
int break_contig(GapIO *io, tg_rec crec, int cpos) {
contig_t *cl;
contig_t *cr;
int cid;
char cname[1024], *cname_end;
int left_end, right_start;
bin_index_t *bin;
int do_comp = 0;
HacheTable *h;
cl = (contig_t *)cache_search(io, GT_Contig, crec);
//contig_dump_ps(io, &cl, "/tmp/tree.ps");
/*
* Our hash table is keyed on sequence record numbers for all sequences
* in all bins spanning the break point. The value is either 0 or 1
* for left/right contig.
*
* The purpose of this hash is to allow us to work out whether a tag
* belongs in the left or right contig, as a tag could start beyond the
* break point but be attached to a sequence before the break point.
*
* Further complicating this is that a tag could be in a smaller bin
* than the sequence as it may not be as long. However we know
* we'll recurse down these in a logical order so we can be sure
* we've already "seen" the sequence that the tag has been
* attached to.
*/
h = HacheTableCreate(1024, HASH_DYNAMIC_SIZE);
strncpy(cname, contig_get_name(&cl), 1000);
cname_end = cname + strlen(cname);
cid = 1;
do {
sprintf(cname_end, "#%d", cid++);
} while (contig_index_query(io, cname) > 0);
if (!(cr = contig_new(io, cname)))
return -1;
cl = cache_rw(io, cl);
cr = cache_rw(io, cr);
if (0 != contig_index_update(io, cname, strlen(cname), cr->rec))
return -1;
printf("Break in contig %"PRIrec", pos %d\n", crec, cpos);
printf("Existing left bin = %"PRIrec", right bin = %"PRIrec"\n",
cl->bin, cr->bin);
cache_incr(io, cl);
cache_incr(io, cr);
bin = get_bin(io, cl->bin);
do_comp = bin->flags & BIN_COMPLEMENTED;
break_contig_recurse(io, h, cl, cr,
contig_get_bin(&cl), cpos, contig_offset(io, &cl),
0, cl->rec, cr->rec, 0, 0);
/* Recompute end positions */
left_end = contig_visible_end(io, cl->rec);
right_start = contig_visible_start(io, cr->rec);
/* Ensure start/end positions of contigs work out */
bin = cache_rw(io, get_bin(io, cr->bin));
//#define KEEP_POSITIONS 1
#ifndef KEEP_POSITIONS
cr->start = 1;
cr->end = cl->end - right_start + 1;
bin->pos -= right_start-1;
#else
cr->start = right_start;
cr->end = cl->end;
#endif
if ((do_comp && !(bin->flags & BIN_COMPLEMENTED)) ||
(!do_comp && (bin->flags & BIN_COMPLEMENTED))) {
bin->flags ^= BIN_COMPLEMENTED;
}
cl->end = left_end;
// remove_redundant_bins(io, cl);
// remove_redundant_bins(io, cr);
printf("Final left bin = %"PRIrec", right bin = %"PRIrec"\n",
cl->bin, cr->bin);
HacheTableDestroy(h, 0);
//if (cl->bin) contig_dump_ps(io, &cl, "/tmp/tree_l.ps");
//if (cr->bin) contig_dump_ps(io, &cr, "/tmp/tree_r.ps");
cache_flush(io);
remove_empty_bins(io, cl->rec);
remove_empty_bins(io, cr->rec);
/* Empty contig? If so remove it completely */
if (cl->bin == 0) {
printf("Removing empty contig %"PRIrec"\n", cl->rec);
contig_destroy(io, cl->rec);
}
if (cr->bin == 0) {
printf("Removing empty contig %"PRIrec"\n", cr->rec);
contig_destroy(io, cr->rec);
}
cache_decr(io, cl);
cache_decr(io, cr);
cache_flush(io);
return 0;
}
int edview_search_consensus(edview *xx, int dir, int strand, char *value) {
int mismatches = 0; /* exact match */
int where = 2; /* consensus */
char *p;
int start, end;
char cons[WIN_WIDTH+1];
int patlen;
char *uppert, *upperb;
int found = 0, at_end = 0;
tg_rec fseq;
int fpos, i, j;
contig_t *c;
/*
* Parse value search string. It optionally includes two extra params
* separated by #. Ie:
* <string>#<N.mismatches>#<where>.
* <where> is 1 for readings, 2 for consensus, 3 for both.
*/
if (p = strchr(value, '#')) {
mismatches = atoi(p+1);
*p = 0;
if (p = strchr(p+1, '#'))
where = atoi(p+1);
}
/* uppercase search string, remove pads, and store fwd/rev copies */
patlen = strlen(value);
depad_seq(value, &patlen, NULL);
if (NULL == (uppert = (char *)xmalloc(patlen + 1)))
return 0;
if (NULL == (upperb = (char *)xmalloc(patlen + 1)))
return 0;
uppert[patlen] = upperb[patlen] = 0;
for (i = patlen-1; i >= 0; i--) {
upperb[i] = uppert[i] = toupper(value[i]);
}
complement_seq(upperb, patlen);
/* Loop */
if (dir) {
start = xx->cursor_apos + (dir ? 1 : -1);
end = start + (WIN_WIDTH-1);
} else {
end = xx->cursor_apos + (dir ? 1 : -1);
start = end - (WIN_WIDTH-1);
}
fpos = xx->cursor_apos;
c = cache_search(xx->io, GT_Contig, xx->cnum);
cache_incr(xx->io, c);
do {
char *ind, *indt = NULL, *indb = NULL;
calculate_consensus_simple(xx->io, xx->cnum, start, end, cons, NULL);
cons[WIN_WIDTH] = 0;
if (dir) {
if (strand == '+' || strand == '=')
indt = pstrstr_inexact(cons, uppert, mismatches, NULL);
if (strand == '-' || strand == '=')
indb = pstrstr_inexact(cons, upperb, mismatches, NULL);
} else {
if (strand == '+' || strand == '=')
indt = prstrstr_inexact(cons, uppert, mismatches, NULL);
if (strand == '-' || strand == '=')
indb = prstrstr_inexact(cons, upperb, mismatches, NULL);
}
if (indt && indb)
ind = MIN(indt, indb);
else if (indt)
ind = indt;
else if (indb)
ind = indb;
else
ind = NULL;
if (ind != NULL) {
if (dir) {
if (fpos <= start + ind-cons) {
found = 1;
fpos = start + ind-cons;
fseq = xx->cnum;
}
} else {
if (fpos >= start + ind-cons) {
found = 1;
fpos = start + ind-cons;
fseq = xx->cnum;
}
}
break;
}
/* Next search region - overlapping by patlen+pads */
if (dir) {
for (i = WIN_WIDTH-1, j = patlen; j && i; i--) {
if (cons[i] != '*')
j--;
}
if (i == 0)
break;
start += i;
end += i;
if (start > c->end)
at_end = 1;
} else {
for (i = 0, j = patlen; j && i < WIN_WIDTH; i++) {
if (cons[i] != '*')
j--;
}
if (i == WIN_WIDTH)
break;
start -= WIN_WIDTH-i;
end -= WIN_WIDTH-i;
if (end < c->start)
at_end = 1;
}
} while (!at_end);
cache_decr(xx->io, c);
if (found) {
edSetCursorPos(xx, fseq == xx->cnum ? GT_Contig : GT_Seq,
fseq, fpos, 1);
}
free(uppert);
free(upperb);
return found ? 0 : -1;
}
/*
* Extends the right hand end of a single contig.
*
* Min_depth is the minimum depth for extension. If lower then even if the
* data matches we'll not extend further.
*
* Match_score (+ve) and mismatch_score (-ve) are accumulated during
* extension to ensure that we don't extend into junk mismatching DNA.
*/
static int contig_extend_single(GapIO *io, tg_rec crec, int dir, int min_depth,
int match_score, int mismatch_score) {
int end;
rangec_t *r;
int nr, i;
contig_t *c;
char cons[CSZ], new_cons[ESZ];
int freqs[ESZ][4], depth[ESZ];
double score, best_score;
int best_pos, nseq;
vmessage("Processing contig #%"PRIrec", %s end\n",
crec, dir ? "left" : "right");
for (i = 0; i < ESZ; i++) {
freqs[i][0] = freqs[i][1] = freqs[i][2] = freqs[i][3] = 0;
depth[i] = 0;
}
c = cache_search(io, GT_Contig, crec);
if (NULL == c) return -1;
cache_incr(io, c);
if (consensus_valid_range(io, crec, NULL, &end) != 0) {
cache_decr(io, c);
return -1;
}
calculate_consensus_simple(io, crec, end-(CSZ-1), end, cons, NULL);
/* Start */
/* Not implemented for now: rev complement and go again! */
/* End */
r = contig_seqs_in_range(io, &c, end, end, 0, &nr);
if (!r) {
cache_decr(io, c);
return -1;
}
for (i = 0; i < nr; i++) {
seq_t *s = cache_search(io, GT_Seq, r[i].rec);
seq_t *sorig = s;
int cstart, cend;
int j, k, slen;
if ((s->len < 0) ^ r[i].comp) {
s = dup_seq(s);
complement_seq_t(s);
}
cstart = r[i].start + s->left-1;
cend = r[i].start + s->right-1;
/* Does cutoff extend to contig end, if so does it match cons? */
if (cend < end) {
int mis = 0, len = 0;
if (end - cend >= CSZ) {
/*
fprintf(stderr,"Skipping #%"PRIrec" due to length of cutoff\n",
r[i].rec);
*/
if (sorig != s)
free(s);
r[i].rec = 0; /* Mark for removal */
continue;
}
for (k = s->right, j = cend+1; j <= end; j++, k++) {
//printf("%d: %c %c\n", j, s->seq[k], cons[j-(end-(CSZ-1))]);
if (s->seq[k] != cons[j-(end-(CSZ-1))])
mis++;
}
len = end - cend;
if (100*mis/len > 5) {
/*
fprintf(stderr, "Skipping #%"PRIrec" due to high disagreement "
"with consensus.\n", r[i].rec);
*/
if (sorig != s)
free(s);
r[i].rec = 0;
continue;
}
}
/* So we got here, let's accumulate extension stats */
slen = ABS(s->len);
for (k = 0, j = end+1 - r[i].start; j < slen && k < ESZ; j++, k++) {
//printf("%d: %c\n", j + r[i].start, s->seq[j]);
if(s->seq[j] == 'N')
continue;
freqs[k][dna_lookup[(uint8_t) s->seq[j]]]++;
depth[k]++;
}
if (sorig != s)
free(s);
}
score = best_score = 0;
best_pos = 0;
for (i = 0; i < ESZ; i++) {
int call, best = 0, j;
double dd;
if (depth[i] < min_depth)
break;
for (j = 0; j < 4; j++) {
if (best < freqs[i][j]) {
best = freqs[i][j];
call = j;
}
}
new_cons[i] = "ACGT"[call];
dd = (double)depth[i];
switch (call) {
case 0:
score += freqs[i][0] / dd;
score -= (freqs[i][1] + freqs[i][2] + freqs[i][3]) / dd;
break;
case 1:
score += freqs[i][1] / dd;
score -= (freqs[i][0] + freqs[i][2] + freqs[i][3]) / dd;
break;
case 2:
score += freqs[i][2] / dd;
score -= (freqs[i][0] + freqs[i][1] + freqs[i][3]) / dd;
break;
case 3:
score += freqs[i][3] / dd;
score -= (freqs[i][0] + freqs[i][1] + freqs[i][2]) / dd;
break;
}
if (best_score <= score) {
best_score = score;
best_pos = i+1;
}
/*
printf("%3d %3d\t%c\t%3d %3d %3d %3d %7.1f\n",
i, depth[i], "ACGT"[call],
freqs[i][0], freqs[i][1], freqs[i][2], freqs[i][3],
score);
*/
}
/* printf("Best score is %f at %d\n", best_score, best_pos); */
/* Extend */
nseq = 0;
if (best_pos > 0) {
int furthest_left = end;
for (i = 0; i < nr; i++) {
seq_t *s;
int r_pos;
int score;
if (r[i].rec == 0)
continue;
s = cache_search(io, GT_Seq, r[i].rec);
s = cache_rw(io, s);
if (furthest_left > r[i].start)
furthest_left = r[i].start;
/*
* end + best_pos is the furthest right we can go, but this
* specific read may not be justified in reaching that far
* if it has too many disagreements.
*/
if ((s->len > 0) ^ r[i].comp) {
int best_r = 0, j, k;
int len = ABS(s->len);
//printf(">%s\t", s->name);
r_pos = 0;
score = 0;
//for (k = s->right, j = 0; j < best_pos && k < len; j++, k++) {
for (k = end - r[i].start + 1, j = 0; j < best_pos && k < len; j++, k++) {
if (new_cons[j] == toupper(s->seq[k])) {
score += match_score;
if (best_r <= score) {
best_r = score;
r_pos = k+1;
}
} else {
score += mismatch_score;
}
//putchar(new_cons[j] == toupper(s->seq[k])
// ? toupper(s->seq[k])
// : tolower(s->seq[k]));
}
//putchar('\n');
if (s->right != r_pos) {
s->right = r_pos;
nseq++;
}
} else {
int best_r = 0, j, k;
//printf("<%s\t", s->name);
r_pos = 0;
score = 0;
//for (k = s->left-2, j = 0; j < best_pos && k >= 0; j++, k--) {
for (k = r[i].end - end - 1, j = 0; j < best_pos && k >= 0; j++, k--) {
char b = complement_base(s->seq[k]);
if (new_cons[j] == b) {
score += match_score;
if (best_r <= score) {
best_r = score;
r_pos = k-1;
}
} else {
score += mismatch_score;
}
//putchar(new_cons[j] == toupper(b)
// ? toupper(b)
// : tolower(b));
}
//putchar('\n');
if (s->left != r_pos+2) {
s->left = r_pos+2;
nseq++;
}
}
}
vmessage(" Extended by %d, adjusting %d sequence clip%s\n",
best_pos, nseq, nseq == 1 ? "" : "s");
bin_invalidate_consensus(io, crec, furthest_left, end + best_pos);
} else {
vmessage(" Unable to extend contig\n");
}
free(r);
cache_decr(io, c);
cache_flush(io);
return 0;
}
