int insert_triple_to_heap(Representation* X_rep, Representation* Y_rep,
int ** x_triple_array, int ** y_triple_array, int x_triple_cnt, int y_triple_cnt, PriorityQueue * heap) {
Triple * triple_array = malloc(y_triple_cnt * sizeof(Triple));
int m;
for (m = 0; m < y_triple_cnt; ++m) triple_array[m].rmsd = 11;
int i, j;
int cnt_x = x_triple_cnt;
int cnt_y = y_triple_cnt;
int ** x_triple_array_local = intmatrix(cnt_x, 3);
int ** y_triple_array_local = intmatrix(cnt_y, 3);
memcpy(*x_triple_array_local, *x_triple_array, sizeof (int) * 3 * cnt_x);
memcpy(*y_triple_array_local, *y_triple_array, sizeof (int) * 3 * cnt_y);
int x_triple[3];
int y_triple[3];
double rmsd;
double cutoff_rmsd = 3.0; /* <<<<<<<<<<<<<<<<< hardcoded */
double q_init[4] = {0.0}; // no change
for (i = 0; i < cnt_x; ++i) {
for (j = 0; j < cnt_y; ++j) {
x_triple[0] = x_triple_array_local[i][0];
x_triple[1] = x_triple_array_local[i][1];
x_triple[2] = x_triple_array_local[i][2];
y_triple[0] = y_triple_array_local[j][0];
y_triple[1] = y_triple_array_local[j][1];
y_triple[2] = y_triple_array_local[j][2];
if (!same_hand_triple(X_rep, x_triple, Y_rep, y_triple, 3)) continue;
if (distance_of_nearest_approach(X_rep, x_triple, Y_rep, y_triple, 3, &rmsd)) continue;
if (rmsd > cutoff_rmsd) continue;
triple_array[j].rmsd = rmsd;
memcpy(&triple_array[j].quat, &q_init, 4 * sizeof (double));
memcpy(&triple_array[j].triple_x, &x_triple, 3 * sizeof (double));
memcpy(&triple_array[j].triple_y, &y_triple, 3 * sizeof (double));
}
}
free(triple_array);
return 0;
}
int init_triples_array(Triples_array * triples_array, int size){
if ( ! (triples_array->hhh_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->hhs_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->hsh_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->hss_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->shh_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->ssh_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->shs_array = intmatrix(size,3)) ) return 0;
if ( ! (triples_array->sss_array = intmatrix(size,3)) ) return 0;
return 1;
}
int process_almt (Options *options, Alignment *alignment) {
int retval;
int process_exons (Options *options, Alignment * alignment);
int protected_positions (Options *options, Alignment * alignment);
/* store the position of exons, and replace them with gaps in the alignment */
process_exons (options, alignment);
/* gaps */
count_gaps (options, alignment);
/* protected positions */
protected_positions (options, alignment);
/*allocate space for various indicators of sequence similarity*/
alignment->seq_dist =
dmatrix ( alignment->number_of_seqs, alignment->number_of_seqs);
if ( !alignment->seq_dist ) return 1;
alignment->aligned_sites =
intmatrix ( alignment->number_of_seqs, alignment->number_of_seqs);
if ( ! alignment->aligned_sites ) return 1;
alignment->identical_sites =
intmatrix ( alignment->number_of_seqs, alignment->number_of_seqs);
if ( ! alignment->identical_sites ) return 1;
alignment->similar_sites =
intmatrix ( alignment->number_of_seqs, alignment->number_of_seqs);
if ( ! alignment->similar_sites ) return 1;
retval = seq_pw_dist (alignment);
if ( retval) return retval;
return 0;
}
int in_group_entropy ( Alignment * alignment, int * similar_to, double **score){
int group, col, seq, ctr, aa;
int **freq, *norm; /* ASCII == 128, ASCII size */
double p, entropy;
if ( ! (freq=intmatrix(alignment->no_groups, ASCII)) ) return 1;
if ( ! (norm=emalloc(alignment->no_groups*sizeof(int))) ) return 1;
for (col = 0; col < alignment->length; col++){
/* find frequencies */
for (group=0; group< alignment->no_groups; group++) {
memset (freq[group], 0, ASCII*sizeof(int));
norm[group] = 0;
}
for (seq=0; seq< alignment->number_of_seqs; seq++){
aa = (int) alignment->sequence[seq][col];
if ( aa == 'X' || aa == 'x' ) continue;
if (similar_to) aa=similar_to[aa];
group = alignment->belongs_to_group[seq];
freq [group][aa]++;
norm [group] ++;
}
/* find entropy */
for (group=0; group< alignment->no_groups; group++) {
entropy = 0.0;
for ( ctr=0; ctr < ASCII; ctr++) {
if ( freq[group][ctr] ) {
p = (double)freq[group][ctr]/norm[group];
entropy -= p*log(p);
}
}
score[group][col] = entropy;
}
}
free_imatrix (freq);
free (norm);
return 0;
}
int clustering ( Protein *protein, int * res_rank, int * int_cvg, double *clustering_score){
int ctr, pos;
int L = protein->length;
int first;
int *selection;
int ** adj_matrix;
double weight, avg, std_dev, z;
double cutoff_dist = CUTOFF_DIST;
int cluster_score (int no_of_res, int *seq, int ** adj_matrix,double *score);
int std_dev_over_S (int L, int M, int ** adj_matrix, double *avg, double * std_dev, int first);
selection = (int *) emalloc (protein->length*sizeof(int));
adj_matrix = intmatrix( L,L);
determine_adj_matrix (adj_matrix, protein->sequence, L, cutoff_dist);
for (ctr=0; ctr < L && int_cvg[ctr]; ctr++ ) {
/* turn coverage into selection array */
for (pos = 0; pos < L; pos ++ ) {
selection[pos] = ( res_rank[pos] <= int_cvg[ctr] );
}
/* find weight */
cluster_score (L, selection, adj_matrix, &weight);
/* find avg and stddev in the set of random picks */
std_dev_over_S (L, int_cvg[ctr], adj_matrix, &avg, &std_dev, first = !ctr);
/* evaluate and store the z-score */
z = (std_dev>1.e-5) ? (weight - avg)/std_dev : 0.0;
clustering_score [ctr] = z;
}
free_matrix ((void **)adj_matrix);
free (selection);
return 0;
}
int output_specs_for_refseq ( Options * optin, Protein * protein, Alignment * alignment,
int * almt2protein, Node * leaf,
double ** score, double cutoff , int ** is_precedent) {
int node_id, pos, no_seqs = alignment->number_of_seqs;
int ctr,number_of_nodes_above_cutoff ;
int ctr1, ctr2, node_id_1, node_id_2;
int node_ctr;
int node_pos_ctr, ** node_related_position;
Node * node_ptr, * first_inner_node;
int * above_cutoff = NULL;
char filename[BUFFLEN];
FILE * fptr;
int print_leaves (FILE * fptr, Node * node);
if ( ! ( above_cutoff = emalloc (no_seqs*sizeof(int) ) ) ) exit (1);
if ( ! ( node_related_position = intmatrix ( no_seqs, alignment->length+1) ) ) exit (1);
sprintf (filename, "%s.qry_specs", options->outname);
fptr = efopen (filename, "w");
if ( !fptr) return 1;
for ( pos=0; pos < alignment->length; pos++) {
number_of_nodes_above_cutoff = 0;
ctr = 0;
for ( node_id= 1; node_id < no_seqs; node_id++ ) {
if ( score[pos][node_id] > cutoff ){
above_cutoff [ctr] = node_id;
ctr ++;
}
}
if ( !ctr ) continue;
above_cutoff [ctr] = -1;
number_of_nodes_above_cutoff = ctr;
for (ctr1=0; above_cutoff [ctr1] >= 0; ctr1++ ) {
node_id_1 = above_cutoff [ctr1];
if ( ! node_id_1 ) continue;
for (ctr2 = ctr1+1; above_cutoff [ctr2] >= 0 ; ctr2++ ) {
node_id_2 = above_cutoff [ctr2];
if ( ! node_id_2 ) continue;
/* go for the highest z */
/* if ( is_precedent[node_id_1][node_id_2] || is_precedent[node_id_2][node_id_1] ) {
number_of_nodes_above_cutoff --;
if ( score[pos][node_id_2] < score[pos][node_id_1] ) {
above_cutoff [ctr2] = 0;
} else {
above_cutoff [ctr1] = 0;
break;
}
}*/
/* alternatively, go for the ancestor */
if ( is_precedent[node_id_1][node_id_2] ) {
number_of_nodes_above_cutoff --;
above_cutoff [ctr2] = 0;
} else if ( is_precedent[node_id_2][node_id_1] ) {
number_of_nodes_above_cutoff --;
above_cutoff [ctr1] = 0;
break;
}
}
}
if ( number_of_nodes_above_cutoff <= 0) {
printf (" %d foul\n", pos);
for (ctr1=0; above_cutoff [ctr1] >= 0; ctr1++ ) {
printf (" ** %d \n", above_cutoff [ctr1] );
}
exit (1);
}
if ( number_of_nodes_above_cutoff > 1 ) {
fprintf (fptr, "%4d %5s:", pos + 1, "discr" );
} else {
fprintf (fptr, "%4d %5s:", pos + 1, "det" );
}
for (ctr1=0; above_cutoff [ctr1] >= 0; ctr1++ ) {
// printf (" ** %d \n", above_cutoff [ctr1] );
node_id_1 = above_cutoff [ctr1];
if ( ! node_id_1 ) continue;
fprintf (fptr, " %4d", node_id_1);
node_ctr = 2*no_seqs-1 - node_id_1;
(leaf+node_ctr) -> marked = 1;
node_related_position[node_id_1][0] ++;
node_pos_ctr = node_related_position[node_id_1][0]; /* use the zeroth pos as a counter */
node_related_position[node_id_1][ node_pos_ctr ] = pos + 1;
}
fprintf (fptr, "\n");
}
fprintf ( fptr, "\n\n");
fprintf ( fptr, "nodes:\n=============\n\n");
first_inner_node = leaf + no_seqs;
for ( node_ctr = 0; node_ctr < no_seqs-1; node_ctr ++ ) {
node_ptr = first_inner_node + node_ctr;
if ( node_ptr->marked ) {
fprintf ( fptr, "\n\t node %4d\n", node_ptr->id );
fprintf ( fptr, "\t positions:");
for ( node_pos_ctr=1; node_pos_ctr<= node_related_position[node_ptr->id][0]; node_pos_ctr++ ) {
fprintf ( fptr, " %4d", node_related_position [node_ptr->id][ node_pos_ctr ]);
}
fprintf (fptr, "\n");
fprintf ( fptr, "\t sequences\n");
print_leaves ( fptr, node_ptr);
}
}
fprintf (fptr, "\n");
free (above_cutoff);
free_matrix ( (void*) node_related_position);
//fclose (fptr);
return 0;
}
int main ( int argc, char * argv[]) {
Options options;
Protein protein;
Alignment * alignment = NULL;
int retval;
int almtctr1, almtctr2;
double **score = NULL, corr, pctg_gaps;
double **clustering_score = NULL;
double *area, *distance;
int **rank_order= NULL,**res_rank=NULL,**int_cvg=NULL ;
int ** correlated = NULL, **almt2prot = NULL, **prot2almt = NULL;
/* command file is required */
if ( argc < 2 ) {
fprintf ( stderr, "Usage: %s <command file>.\n", argv[0]);
exit (0);
}
retval = read_cmd_file ( argv[1], &options);
if (retval) exit(retval);
retval = logger (&options, INTRO, "");
if (retval) exit(retval);
/*******************************************/
/* */
/* PDB input */
/* */
/*******************************************/
if ( ! options.pdbname[0]) {
fprintf (stderr, "%s cannot work without structure (cmd file was %s).\n",
argv[0], argv[1]);
exit (1);
} else {
/* warn if no chain given */
if ( !options.chain) {
retval = logger (&options, WARN, "No chain specified. Using the first one.");
if ( retval) exit (1);
}
if (retval) exit(retval);
/* read in the structure */
retval = read_pdb (options.pdbname, &protein, options.chain);
if (retval) exit(retval);
}
/*******************************************/
/* */
/* alignment scoring */
/* */
/*******************************************/
if ( ! ( alignment = emalloc ( options.no_of_alignments*sizeof(Alignment)) )) return 1;
if ( ! ( score = emalloc ( options.no_of_alignments*sizeof(double*)) )) return 1;
if ( ! ( rank_order = emalloc ( options.no_of_alignments*sizeof(int*)) )) return 1;
if ( ! ( clustering_score = emalloc ( options.no_of_alignments*sizeof(double*)) )) return 1;
if ( ! ( res_rank = emalloc ( options.no_of_alignments*sizeof(int*)) )) return 1;
if ( ! ( int_cvg = emalloc ( options.no_of_alignments*sizeof(int*)) )) return 1;
if ( ! ( almt2prot = emalloc ( options.no_of_alignments*sizeof(int*)) )) return 1;
if ( ! ( prot2almt = emalloc ( options.no_of_alignments*sizeof(int*)) )) return 1;
if ( ! ( area = emalloc ( options.no_of_alignments*sizeof(double)) )) return 1;
if ( ! ( distance = emalloc ( options.no_of_alignments*sizeof(double)) )) return 1;
printf ( "\t%8s %20s %8s %8s %8s \n", "almt#", "name ", "<dist to qry>", "%gaps", "area");
for ( almtctr1 = 0; almtctr1 < options.no_of_alignments; almtctr1++) {
/* read in the alignment */
retval = read_clustalw (options.almtname[almtctr1], alignment + almtctr1);
if (retval) exit(retval);
/* pairwise distances btw the seqs */
retval = seq_pw_dist (alignment+almtctr1);
if ( retval) return retval;
/* average dist to the query in this alignment: */
distance[almtctr1] = avg_dist_to_special (&options, alignment + almtctr1);
/* percentage of gaps in the alignment: */
pctg_gaps = (double) alignment->total_gaps/ ( (alignment+almtctr1)->length*(alignment+almtctr1)->number_of_seqs);
/* make the residue scoring array */
score[almtctr1] = emalloc ( alignment[almtctr1].length*sizeof(double));
/* fill in the score array */
scoring (&options, alignment+almtctr1, score[almtctr1]);
/* translate the scoring into rank order */
rank_order[almtctr1] = emalloc ( alignment[almtctr1].length*sizeof(int));
score2rank (score[almtctr1], rank_order[almtctr1], alignment[almtctr1].length);
/* mapping between the protein and the alignment almtctr1 */
if ( ! (almt2prot[almtctr1] = (int *) emalloc (alignment[almtctr1].length*sizeof(int))) )exit (1);
if ( ! (prot2almt[almtctr1] = (int *) emalloc (protein.length*sizeof(int))) )exit (1);
retval = struct_almt_mapping (&protein, alignment+almtctr1, options.query, prot2almt[almtctr1], almt2prot[almtctr1]);
if (retval) exit(retval);
/* find coverage info implied by the scoring array */
if ( ! (res_rank[almtctr1] = (int*) emalloc (protein.length*sizeof(int))) ) exit (1);
if ( ! (int_cvg[almtctr1] = (int*) emalloc (protein.length*sizeof(int))) ) exit (1);
coverage ( &protein, almt2prot[almtctr1], score[almtctr1], alignment[almtctr1].length,
res_rank[almtctr1], int_cvg[almtctr1] );
/*clustering score*/
clustering_score[almtctr1] = (double*) emalloc (protein.length*sizeof(double));
if (!clustering_score[almtctr1]) exit(retval);
clustering ( &protein, res_rank[almtctr1], int_cvg[almtctr1], clustering_score[almtctr1]);
/* cumulative clustering score*/
area[almtctr1] = area_over_coverage (int_cvg[almtctr1], clustering_score[almtctr1], protein.length);
printf ( "\t %4d %20s %8.3lf %8.3lf %8.3lf \n",
almtctr1, options.almtname[almtctr1], distance[almtctr1], pctg_gaps, area[almtctr1]);
}
/* find the table of correlations */
if ( ! (correlated = intmatrix ( options.no_of_alignments, options.no_of_alignments) ) ) return 1;
for ( almtctr1 = 0; almtctr1 < options.no_of_alignments -1; almtctr1++) {
correlated[almtctr1][almtctr1] = 1;
for ( almtctr2 = almtctr1+1; almtctr2 < options.no_of_alignments; almtctr2++) {
if ( alignment[almtctr1].length != alignment[almtctr2].length ) {
fprintf ( stderr, "Error alignments in the files %s and %s ",
options.almtname[almtctr1], options.almtname[almtctr2]);
fprintf ( stderr, "seem to be of unequal length: %d and %d.\n",
alignment[almtctr1].length , alignment[almtctr2].length);
return 1;
}
corr = spearman ( rank_order[almtctr1], rank_order[almtctr2], alignment[almtctr1].length );
printf ( " %3d %3d %8.4lf\n", almtctr1, almtctr2, corr);
correlated[almtctr1][almtctr2] = ( corr > 0.9 );
}
}
/* find corelated clusters (of sequence selections)*/
{
int *cluster_count_per_size;
int no_of_clusters;
int max_size, secnd_max_size , ** cluster;
int size = options.no_of_alignments;
int i,j;
double dist, ar, max_area, dist_at_max_area;
double min_dist_at_max_area, min_dist, max_area_at_min_dist;
int almt_no, min_dist_almt;
int cluster_counter (int no_of_things, int *neighbors[],
int cluster_count_per_size[], int * no_of_clusters,
int * max_size, int * secnd_max_size , int * cluster[]);
if ( ! ( cluster_count_per_size = emalloc (size*sizeof(int)))) return 1;
if ( ! (cluster = intmatrix ( size+1, size+1) ) ) return 1;
retval = cluster_counter (size, correlated, cluster_count_per_size, &no_of_clusters,
& max_size, &secnd_max_size , cluster);
if ( retval ) return 1;
printf ( "number of clusters: %d \n", no_of_clusters);
for (i=0; i<=size; i++ ) {
if ( ! cluster[i][0] ) continue;
if ( !i ) {
printf ( "\t isolated:\n");
} else {
printf ("\t cluster size: %3d \n", cluster[i][0]);
}
for (j=1; j <= cluster[i][0]; j++ ) {
printf ( "%3d ", cluster[i][j] );
}
printf ( "\n");
}
/* which cluster is the closest to the singled out sequence ("special") */
min_dist_at_max_area = dist_at_max_area = 10;
max_area_at_min_dist = min_dist = -10;
min_dist_almt = -1;
for (i=0; i<=size; i++ ) {
if ( ! cluster[i][0] ) continue;
max_area = -100;
almt_no = dist_at_max_area = -1;
for (j=1; j <= cluster[i][0]; j++ ) {
dist = distance[cluster[i][j]] ;
ar = area[cluster[i][j]] ;
if ( max_area < ar ) {
max_area = ar;
dist_at_max_area = dist;
almt_no = cluster[i][j];
}
}
if ( almt_no < 0 ) {
fprintf ( stderr, "Error selecting the alignment (1)\n");
exit (1);
}
if ( min_dist_at_max_area > dist_at_max_area ) {
min_dist = dist_at_max_area;
max_area_at_min_dist = max_area;
min_dist_almt = almt_no;
}
}
if ( min_dist_almt < 0 ) {
fprintf ( stderr, "Error selecting the alignment (2)\n");
exit (1);
}
printf ( "choosing alignment %d %s (distance: %5.3f area: %6.3f)\n",
min_dist_almt, options.almtname[min_dist_almt], min_dist, max_area_at_min_dist);
free (cluster_count_per_size);
free_matrix ( (void **) cluster);
}
free (score);
logger ( &options, NOTE, "");
return 0;
}
int output_specs_score ( Options *options, Protein * protein, Alignment * alignment,
int *almt2prot, Node * leaf,
double ** score, double ** complement_score, double ** p_value,
double **probability, double **overlap,
double cutoff, int ** is_precedent) {
# if 0
char *base_filename = options ->outname;
int node_id, pos, no_seqs = alignment->number_of_seqs;
int node_id_2;
int node_ctr,* print, first;
int ** related;
int int_max_gaps = MAX_GAPS*no_seqs;
Node *root;
int refseq_number, refseq_2_number = -1;
char filename[BUFFLEN];
char in_qry2;
FILE * fptr;
int print_leaves (FILE * fptr, Node * node);
int recursive_cleanup (Node *node, int * related, double * probability);
if ( ! ( related = intmatrix (alignment->length+1, no_seqs ) ) ) exit (1);
if ( ! ( print = emalloc (no_seqs*sizeof(int) ) ) ) exit (1);
printf ("in output\n");
#if 1
sprintf (filename, "%s.specs", base_filename);
fptr = efopen (filename, "w");
if ( !fptr) return 1;
# else
fptr = stdout;
# endif
if ( ! options->refseq ) {
fprintf ( stderr, "Please define refseq.\n");
exit (0);
}
/* locate the refseq */
refseq_number = find_refseq (alignment, options->refseq);
if ( options->qry2[0]) refseq_2_number = find_refseq (alignment, options->qry2);
/* associate nodes and positions */
for ( node_id = 2; node_id < no_seqs; node_id++ ) {
node_ctr = 2*no_seqs -1 - node_id;
if ( (leaf+node_ctr)-> number_of_leaves < MIN_NO_LEAVES) continue;
for ( pos=0; pos < alignment->length; pos++) {
if ( alignment->column_gaps[pos] > int_max_gaps) continue;
if ( overlap[pos][node_id] < MAX_OVERLAP && probability[pos][node_id] <= MAX_PROB) {
related[pos][node_id] = 1;
}
}
}
/* recursive cleanup (so parents and children do not appear for the same reason */
for ( pos=0; pos < alignment->length; pos++) {
if ( alignment->column_gaps[pos] > int_max_gaps ) continue;
recursive_cleanup ( root = leaf+2*no_seqs-1 - 1, related[pos], probability[pos] ) ;
}
/* which nodes should be printed? */
for ( node_id = 2; node_id < no_seqs; node_id++ ) {
for ( pos=0; pos < alignment->length; pos++) {
if ( alignment->column_gaps[pos] > int_max_gaps ) continue;
if ( related[pos][node_id] ) {
print[node_id] = 1;
break;
}
}
}
for ( node_id = 2; node_id < no_seqs; node_id++ ) {
node_ctr = 2*no_seqs-1 - node_id;
if ( ! print[node_id] ) continue;
fprintf ( fptr, "\n\nnode %d\n=============\n", node_id);
fprintf ( fptr, " %4s %4s %4s %4s %5s %5s %5s %5s other nodes\n",
"pos", "pdbid", "qry1", "qry2", "entr", "c_entr", "prob", "ovlp");
for ( pos=0; pos < alignment->length; pos++) {
if ( ! related[pos][node_id] ) continue;
in_qry2 = (options->qry2[0]) ? alignment->sequence[refseq_2_number][pos]: '-';
fprintf ( fptr, " %4d %4s %1c %1c %5.1lf %5.1lf %5.1le %5.2lf ",
pos+1, protein->sequence[ almt2prot[pos]].pdb_id,
alignment->sequence[refseq_number][pos],
in_qry2,
score[pos][node_id], complement_score[pos][node_id],
probability[pos][node_id], overlap[pos][node_id] );
/* other nodes which have this same pos as discriminant */
first = 1;
for ( node_id_2 = 2; node_id_2 < no_seqs; node_id_2 ++ ) {
if ( node_id_2 == node_id ) continue;
if ( related[pos][node_id_2] ) {
if ( ! first ) {
fprintf ( fptr, ",");
}
fprintf ( fptr, "%d",node_id_2);
first = 0;
}
}
fprintf ( fptr, "\n");
}
}
fclose (fptr);
/* print out leaves */
sprintf (filename, "%s.leaves", base_filename);
fptr = efopen (filename, "w");
if ( !fptr) return 1;
for ( node_id = 2; node_id < no_seqs; node_id++ ) {
if ( ! print[node_id] ) continue;
node_ctr = 2*no_seqs -1 - node_id;
fprintf ( fptr, "\n\nnode %d\n=============\n", node_id);
print_leaves ( fptr, leaf+node_ctr);
fprintf (fptr, "\n");
}
fclose (fptr);
# endif
return 0;
}
int find_best_triples_exhaustive_parallel(Representation* X_rep, Representation* Y_rep, int no_top_rmsd,
double * best_rmsd, int ** best_triple_x, int ** best_triple_y,
double **best_quat) {
// initialization of global array of values
no_top_rmsd = TOP_RMSD;
// printf("%d\n", no_top_rmsd);
// printf("proba %d %lf\n", X_rep->N_full, X_rep->cm[0][0]);
double ** best_quat_array = dmatrix(no_top_rmsd * NUM_THREADS, 4);
int ** best_triple_x_array = intmatrix(no_top_rmsd * NUM_THREADS, 3);
int ** best_triple_y_array = intmatrix(no_top_rmsd * NUM_THREADS, 3);
double * best_rmsd_array = (double *) malloc(no_top_rmsd * NUM_THREADS * sizeof (double));
int cnt;
for (cnt = 0; cnt < NUM_THREADS * no_top_rmsd; ++cnt) {
best_rmsd_array[cnt] = BAD_RMSD + 1;
best_triple_x_array[cnt][0] = -1;
}
omp_set_num_threads(NUM_THREADS);
#pragma omp parallel
{
int top_ctr, i, j, k, l, n, m;
int myid = omp_get_thread_num();
double ** best_quat_local = dmatrix(no_top_rmsd, 4);
int ** best_triple_x_local = intmatrix(no_top_rmsd, 3);
int ** best_triple_y_local = intmatrix(no_top_rmsd, 3);
double * best_rmsd_local = (double *) malloc(no_top_rmsd * sizeof (double));
double **x = X_rep->full; // no change
int * x_type = X_rep->full_type; // no change
int NX = X_rep->N_full; // no change
double **y = Y_rep->full;
int * y_type = Y_rep->full_type;
int NY = Y_rep->N_full;
int x_triple[3], y_triple[3];
int chunk;
double cutoff_rmsd = 3.0; /* <<<<<<<<<<<<<<<<< hardcoded */
double rmsd; //
double q_init[4] = {0.0}; // no change
double ** cmx = X_rep->cm; // no change
double ** cmy = Y_rep->cm; // no change
double threshold_dist = THRESHOLD;
/***************************************/
/* find reasonable triples of SSEs */
/* that correspond in type */
/* and can be mapped onto each other */
/***************************************/
for (top_ctr = 0; top_ctr < no_top_rmsd; top_ctr++) {
best_rmsd_local[top_ctr] = BAD_RMSD + 1;
best_triple_x_local[top_ctr][0] = -1;
}
/*
* Exhaustive search through a 6D space - ugly code
* Parallelization
*/
#pragma omp for
for (i = 0; i < NX; ++i) {
for (j = 0; j < NY - 2; ++j) {
if (x_type[i] != y_type[j]) continue;
for (k = 0; k < NX; ++k) {
if (k == i) continue;
if (two_point_distance(cmx[i], cmx[k]) > THRESHOLD) continue;
for (l = j + 1; l < NY - 1; ++l) {
if (x_type[k] != y_type[l]) continue;
if (two_point_distance(cmy[j], cmy[l]) > THRESHOLD) continue;
for (m = 0; m < NX; ++m) {
if (m == k || m == i) continue;
if (two_point_distance(cmx[i], cmx[m]) > THRESHOLD) continue;
if (two_point_distance(cmx[k], cmx[m]) > THRESHOLD) continue;
for (n = l + 1; n < NY; ++n) {
if (x_type[m] != y_type[n]) continue;
if (two_point_distance(cmy[j], cmy[n]) > THRESHOLD) continue;
if (two_point_distance(cmy[l], cmy[n]) > THRESHOLD) continue;
x_triple[0] = i;
y_triple[0] = j;
x_triple[1] = k;
y_triple[1] = l;
x_triple[2] = m;
y_triple[2] = n;
if (!same_hand_triple(X_rep, x_triple, Y_rep, y_triple, 3)) continue;
if (distance_of_nearest_approach(X_rep, x_triple,
Y_rep, y_triple, 3, &rmsd)) continue;
if (rmsd > cutoff_rmsd) continue;
//if (opt_quat(x, NX, x_triple, y, NY, y_triple, 3, q_init, &rmsd)) continue;
for (top_ctr = 0; top_ctr < no_top_rmsd; top_ctr++) {
// insertion of a new values in arrays keeping arrays sorted
if (rmsd <= best_rmsd_local[top_ctr]) {
chunk = no_top_rmsd - top_ctr - 1;
if (chunk) {
memmove(best_rmsd_local + top_ctr + 1,
best_rmsd_local + top_ctr, chunk * sizeof(double));
memmove(best_quat_local[top_ctr + 1],
best_quat_local[top_ctr], chunk * 4 * sizeof(double));
memmove(best_triple_x_local[top_ctr + 1],
best_triple_x_local[top_ctr], chunk * 3 * sizeof (int));
memmove(best_triple_y_local[top_ctr + 1],
best_triple_y_local[top_ctr], chunk * 3 * sizeof (int));
}
best_rmsd_local[top_ctr] = rmsd;
memcpy(best_quat_local[top_ctr], q_init, 4 * sizeof (double));
memcpy(best_triple_x_local[top_ctr], x_triple, 3 * sizeof (int));
memcpy(best_triple_y_local[top_ctr], y_triple, 3 * sizeof (int));
break;
}
}
}
}
}
}
}
// printf("%d\n", i);
// each thread copies values to global arrays in accordance with its thread id
// printf("myid: %d\n", myid);
memcpy(*(best_quat_array + myid * no_top_rmsd), *(best_quat_local), no_top_rmsd * 4 * sizeof (double));
memcpy(*(best_triple_y_array + myid * no_top_rmsd), *(best_triple_y_local), no_top_rmsd * 3 * sizeof (int));
memcpy(*(best_triple_x_array + myid * no_top_rmsd), *(best_triple_x_local), no_top_rmsd * 3 * sizeof (int));
memcpy(best_rmsd_array + myid*no_top_rmsd, best_rmsd_local, no_top_rmsd * sizeof (double));
}
free_dmatrix(best_quat_local);
free_imatrix(best_triple_x_local);
free_imatrix(best_triple_y_local);
free(best_rmsd_local);
// parallel sort of elements of arrays
sortTriplets(best_triple_x_array, best_triple_y_array, best_rmsd_array, best_quat_array, no_top_rmsd);
}
//
/*
memcpy(*best_quat, *best_quat_array, no_top_rmsd * 4 * sizeof(double));
*/
memcpy(*best_triple_y, *best_triple_y_array, no_top_rmsd * 3 * sizeof (int));
memcpy(*best_triple_x, *best_triple_x_array, no_top_rmsd * 3 * sizeof (int));
memcpy(best_rmsd, best_rmsd_array, no_top_rmsd * sizeof (double));
free_dmatrix(best_quat_array);
free_imatrix(best_triple_x_array);
free_imatrix(best_triple_y_array);
free(best_rmsd_array);
return 0;
}
int complement_match (Representation* X_rep, Representation* Y_rep,
Map * map, int map_max,
int * map_ctr, int * map_best, int best_max, int parent_map){
Penalty_parametrization penalty_params; /* for SW */
double **x = X_rep->full;
int * x_type = X_rep->full_type;
int NX = X_rep->N_full;
double **y = Y_rep->full;
int * y_type = Y_rep->full_type;
int NY = Y_rep->N_full;
double F_effective = 0.0;
double F_current;
double q[4] = {0.0}, q_init[4] = {0.0};
double **x_rotated = NULL;
double **tr_x_rotated = NULL;
double **R;
double z_scr = 0.0, *z_best;
double avg, avg_sq, stdev;
double alpha = options.alpha;
double rmsd, best_rmsd[TOP_RMSD];
double **best_quat;
double cutoff_rmsd = 3.0; /* <<<<<<<<<<<<<<<<< hardcoded */
int *x_type_fudg, *y_type_fudg;
int *anchor_x, *anchor_y, no_anchors;
int no_top_rmsd = TOP_RMSD, chunk;
int x_ctr, y_ctr, top_ctr;
int **best_triple_x;
int **best_triple_y;
int x_triple[3], y_triple[3];
int retval, done = 0;
int best_ctr;
int i, j;
int t;
int smaller;
int my_map_ctr;
int stored_new;
int * x2y, map_unstable;
//time_t time_now, time_start;
int cull_by_dna (Representation * X_rep, int *set_of_directions_x,
Representation * Y_rep, int *set_of_directions_y,
int set_size, Map *map, double cutoff_rmsd);
int distance_of_nearest_approach (Representation * X_rep, int *set_of_directions_x,
Representation * Y_rep, int *set_of_directions_y,
int set_size, double * rmsd_ptr);
int same_hand_triple (Representation * X_rep, int *set_of_directions_x,
Representation * Y_rep, int *set_of_directions_y, int set_size);
int find_map (Penalty_parametrization * params, Representation *X_rep, Representation *Y_rep,
double ** R, double alpha, double * F_effective, Map *map,
int *anchor_x, int * anchor_y, int anchor_size );
int find_next_triple (double **X, double **Y,
int *x_type, int *y_type, int NX, int NY,
int *x_triple, int *y_triple);
int gradient_descent (int first_call, double alpha,
double **x, int * x_type, int NX,
double **y, int * y_type, int NY,
double *q_best, double *F_best_ptr) ;
int map_quality_metrics (Representation *X_rep, Representation *Y_rep,
double ** tr_x_rotated, Map * map, int *reasonable_angle_ct);
int monte_carlo (double alpha,
double **x, int * x_type, int NX,
double **y, int * y_type, int NY,
double *q_best, double *F_best_ptr);
int opt_quat (double ** x, int NX, int *set_of_directions_x,
double ** y, int NY, int *set_of_directions_y,
int set_size, double * q, double * rmsd);
int qmap (double *x0, double *x1, double *y0, double *y1, double * quat);
int store_sorted (Map * map, int NX, int NY, int *map_best, int map_max,
double * z_best, int best_ctr,
double z_scr, int my_map_ctr, int *stored);
map_best[0] = -1; /* it is the end-of-array flag */
if ( *map_ctr >= map_max ) {
fprintf (stderr, "Map array undersized.\n");
exit (1);
}
smaller = (NX <= NY) ? NX : NY;
/***********************/
/* memory allocation */
/***********************/
if ( ! (R=dmatrix(3,3) ) ) return 1; /* compiler is bugging me otherwise */
if ( ! (x_rotated = dmatrix (NX,3)) ) return 1;
if ( ! (tr_x_rotated = dmatrix (NX,3)) ) return 1;
if ( ! (best_quat = dmatrix (no_top_rmsd,4)) ) return 1;
if ( ! (best_triple_x = intmatrix (no_top_rmsd,3)) ) return 1;
if ( ! (best_triple_y = intmatrix (no_top_rmsd,3)) ) return 1;
if ( ! (z_best = emalloc(NX*NY*sizeof(double) )) ) return 1;
if ( ! (x_type_fudg = emalloc(NX*sizeof(int) )) ) return 1;
if ( ! (y_type_fudg = emalloc(NY*sizeof(int) )) ) return 1;
if ( ! (anchor_x = emalloc(NX*sizeof(int) )) ) return 1;
if ( ! (anchor_y = emalloc(NY*sizeof(int) )) ) return 1;
penalty_params.custom_gap_penalty_x = NULL;
penalty_params.custom_gap_penalty_y = NULL;
//if ( ! (penalty_params.custom_gap_penalty_x = emalloc(NX*sizeof(double) )) ) return 1;
//if ( ! (penalty_params.custom_gap_penalty_y = emalloc(NY*sizeof(double) )) ) return 1;
/***********************/
/***********************/
/* expected quantities */
/***********************/
avg = avg_sq = stdev = 0.0;
//if (options.postprocess) {
if (0) {
if (F_moments (x, x_type, NX, y, y_type, NY, alpha, &avg, &avg_sq, &stdev)) return 1;
}
/***********************/
/***********************/
/* initialization */
/***********************/
best_ctr = 0;
penalty_params.gap_opening = options.gap_open;
penalty_params.gap_extension = options.gap_extend;
penalty_params.endgap = options.endgap;
penalty_params.endgap_special_treatment = options.use_endgap;
/***********************/
/***************************************/
/* find reasonble triples of SSEs */
/* that correspond in type */
/* and can be mapped onto each other */
/***************************************/
for (top_ctr=0; top_ctr<no_top_rmsd; top_ctr++) {
best_rmsd[top_ctr] = BAD_RMSD+1;
best_triple_x[top_ctr][0] = -1;
}
for (x_ctr=0; x_ctr < NX-2 && !done; x_ctr++) {
for (y_ctr=0; y_ctr < NY-2 && !done; y_ctr++) {
if ( y_type[y_ctr] != x_type[x_ctr] ) continue;
x_triple[0] = x_ctr;
y_triple[0] = y_ctr;
if (find_next_triple (x, y, x_type, y_type,
NX, NY, x_triple, y_triple) ){
continue;
}
if ( x_triple[1] < 0 || x_triple[2] < 0 ) continue;
if ( y_triple[1] < 0 || y_triple[2] < 0 ) continue;
/* do these three have kind-of similar layout in space?*/
/* is handedness the same? */
if ( ! same_hand_triple ( X_rep, x_triple, Y_rep, y_triple, 3)) continue;
/* are distances comparab;e? */
if (distance_of_nearest_approach ( X_rep, x_triple,
Y_rep, y_triple, 3, &rmsd)) continue;
if ( rmsd > cutoff_rmsd) continue;
/* find q_init that maps the two triples as well as possible*/
if ( opt_quat ( x, NX, x_triple, y, NY, y_triple, 3, q_init, &rmsd)) continue;
for (top_ctr=0; top_ctr<no_top_rmsd; top_ctr++) {
if ( rmsd <= best_rmsd[top_ctr] ) {
chunk = no_top_rmsd - top_ctr -1;
if (chunk) {
memmove (best_rmsd+top_ctr+1, best_rmsd+top_ctr, chunk*sizeof(double));
memmove (best_quat[top_ctr+1],
best_quat[top_ctr], chunk*4*sizeof(double));
memmove (best_triple_x[top_ctr+1],
best_triple_x[top_ctr], chunk*3*sizeof(int));
memmove (best_triple_y[top_ctr+1],
best_triple_y[top_ctr], chunk*3*sizeof(int));
}
best_rmsd[top_ctr] = rmsd;
memcpy (best_quat[top_ctr], q_init, 4*sizeof(double));
memcpy (best_triple_x[top_ctr], x_triple, 3*sizeof(int));
memcpy (best_triple_y[top_ctr], y_triple, 3*sizeof(int));
break;
}
}
}
}
# if 0
for (top_ctr=0; top_ctr<no_top_rmsd; top_ctr++) {
if ( best_rmsd[top_ctr] > BAD_RMSD ) break;
printf (" %3d %8.3lf ", top_ctr, best_rmsd[top_ctr]);
vec_out ( best_quat[top_ctr], 4, "quat: ");
for (t=0; t<3; t++ ) {
printf ("\t %3d %3d \n", best_triple_x[top_ctr][t]+1, best_triple_y[top_ctr][t]+1 );
}
}
exit (1);
# endif
/*********************************************/
/* main loop */
/*********************************************/
for (top_ctr=0; top_ctr<no_top_rmsd; top_ctr++) {
if ( best_rmsd[top_ctr] > BAD_RMSD ) break;
quat_to_R (best_quat[top_ctr], R);
rotate (x_rotated, NX, R, x);
F_current = F( y, y_type, NY, x_rotated, x_type, NX, alpha);
# if 0
printf ("\n***********************************\n");
printf (" %3d %8.3lf %8.3lf ", top_ctr, best_rmsd[top_ctr], F_current);
vec_out ( best_quat[top_ctr], 4, "quat: ");
for (t=0; t<3; t++ ) {
printf ("\t %3d %3d \n", best_triple_x[top_ctr][t]+1, best_triple_y[top_ctr][t]+1 );
}
# endif
/* find map which uses the 2 triples as anchors */
no_anchors = 3;
find_map (&penalty_params, X_rep, Y_rep, R, alpha, &F_effective, map + (*map_ctr),
best_triple_x[top_ctr], best_triple_y[top_ctr], no_anchors);
x2y = ( map + (*map_ctr) ) ->x2y;
map_unstable = 0;
for (t=0; t<3; t++ ) {
if ( x2y[best_triple_x[top_ctr][t]] != best_triple_y[top_ctr][t] ) {
map_unstable = 1;
}
}
if ( map_unstable) continue;
/* dna here is not DNA but "distance of nearest approach" */
cull_by_dna ( X_rep, best_triple_x[top_ctr],
Y_rep, best_triple_y[top_ctr], 3, map + (*map_ctr), cutoff_rmsd );
//printf ("map after culling by dna:\n");
//print_map (stdout, map+ (*map_ctr), NULL, NULL, NULL, NULL, 1);
/* monte that optimizes the aligned vectors only */
for (i=0; i<NX; i++) {
x_type_fudg[i] = JACKFRUIT;
}
for (j=0; j<NY; j++) {
y_type_fudg[j] = JACKFRUIT*2;
}
no_anchors = 0;
for (i=0; i<NX; i++) {
j = (map+(*map_ctr))->x2y[i];
if (j < 0 ) continue;
x_type_fudg[i] = x_type[i];
y_type_fudg[j] = y_type[j];
anchor_x[no_anchors] = i;
anchor_y[no_anchors] = j;
no_anchors ++;
}
if ( opt_quat ( x, NX, anchor_x, y, NY, anchor_y, no_anchors, q, &rmsd)) continue;
retval = monte_carlo ( alpha, x, x_type_fudg, NX,
y, y_type_fudg, NY, q, &F_current);
if (retval) return retval;
if (options.postprocess) {
z_scr = stdev ? (F_current - avg)/stdev : 0.0;
} else {
z_scr = 0.0;
}
quat_to_R (q, R);
/* store_image() is waste of time, but perhaps not critical */
store_image (X_rep, Y_rep, R, alpha, map + (*map_ctr));
map_assigned_score ( X_rep, map + (*map_ctr));
//printf ("map %2d assigned score: %8.3lf z_score: %8.3lf \n\n",
// *map_ctr+1, (map + (*map_ctr)) -> assigned_score, z_scr);
/* store the map that passed all the filters down to here*/
my_map_ctr = *map_ctr;
map[my_map_ctr].F = F_current;
map[my_map_ctr].avg = avg;
map[my_map_ctr].avg_sq = avg_sq;
map[my_map_ctr].z_score = z_scr;
memcpy ( map[my_map_ctr].q, q, 4*sizeof(double) );
/* recalculate the assigned score*/
//if (top_ctr==24) exit (1);
/************************/
/* store sorted */
/************************/
/* find the place for the new z-score */
store_sorted (map, NX, NY, map_best, map_max,
z_best, best_ctr, -map[my_map_ctr].assigned_score, my_map_ctr, &stored_new);
if ( stored_new ) { /* we want to keep this map */
(*map_ctr) ++;
best_ctr++;
} /* otherwise this map space is reusable */
/* is this pretty much as good as it can get ? */
if ( fabs (map[my_map_ctr].assigned_score - smaller)
< options.tol ) done = 1;
}
map_best[best_ctr] = -1;
/******************************************************/
/* look for the sub-map of a couple of best hits */
/******************************************************/
/* initialization:*/
map_consistence ( NX, NY, NULL, NULL, NULL, NULL, NULL);
best_ctr = 0;
while ( map_best[best_ctr] > -1 ) {
best_ctr ++;
}
//exit (1);
if (best_ctr) {
int nr_maps = (best_ctr<options.number_maps_cpl)?
best_ctr : options.number_maps_cpl;
int best_i;
int consistent;
double z;
double total_assigned_score, score, best_score = -100;
double gap_score;
for (i=0; i<nr_maps; i++) { /* look for the complement */
best_i = map_best[i];
/*intialize the (list of) submatch map(s) */
if ( !map[best_i].submatch_best) {
/* for now look for a single map only */
/* TODO - would it be worth any to look at more maps?*/
int map_max = 1;
map[best_i].submatch_best = emalloc (map_max*sizeof(int) );
if (! map[best_i].submatch_best) return 1;
}
map[best_i].submatch_best[0] = -1;
map[best_i].score_with_children = 0;
map[best_i].compl_z_score = 0;
for (j=0; j<best_ctr; j++) {
if (i==j) continue;
map_complementarity ( map+best_i, map + map_best[j], &z);
map_consistence ( NX, NY, map+best_i, map + map_best[j],
&total_assigned_score, &gap_score, NULL);
consistent = ( (map+best_i)->assigned_score < total_assigned_score
&& (map + map_best[j])->assigned_score
< total_assigned_score);
if ( consistent ) {
score = total_assigned_score;
if ( score > best_score ) {
best_score = score;
map[best_i].submatch_best[0] = map_best[j];
map[best_i].score_with_children = total_assigned_score;
map[best_i].compl_z_score = z;
}
}
}
}
}
/**********************/
/* garbage collection */
gradient_descent (1, 0.0, NULL, NULL, 0,
NULL, NULL, 0, NULL, NULL);
free_dmatrix (R);
free_dmatrix (x_rotated);
free_dmatrix (tr_x_rotated);
free_dmatrix (best_quat);
free_imatrix (best_triple_x);
free_imatrix (best_triple_y);
free (z_best);
free (x_type_fudg);
free (y_type_fudg);
free (anchor_x);
free (anchor_y);
if (penalty_params.custom_gap_penalty_x) free (penalty_params.custom_gap_penalty_x);
if (penalty_params.custom_gap_penalty_y) free (penalty_params.custom_gap_penalty_y);
/*********************/
return 0;
}
