static void block_init(Edge *e, Block *b,
struct dyStack *genes, struct dyStack *scores,
bool *candidates, const int cand_threshold,
int *components, struct dyStack *allincluster)
{
int i,j=0,score,top;
int cnt = 0, cnt_all=0, col_num=0;
continuous cnt_ave=0, row_all = rows;
double pvalue=0;
int max_cnt, max_i;
int *arr_rows, *arr_rows_b;
AllocArray(arr_rows, rows);
AllocArray(arr_rows_b, rows);
bool *colcand;
AllocArray(colcand, cols);
for (i=0; i< cols; i++)
colcand[i] = FALSE;
discrete *g1, *g2;
discrete *sub_array, col_array[2], col_all[rows];
continuous KL_score[cols],KL_score_c=0, KL_score_r=0;
g1 = arr_c[dsItem(genes,0)];
g2 = arr_c[dsItem(genes,1)];
/*update intial colcand*/
for (i=0; i< cols; i++)
{
if ((g1[i] == g2[i])&&(symbols[g1[i]]!=0))
{
colcand[i] = TRUE;
for (j=0;j<rows;j++)
col_all[j]=arr_c[j][i];
col_array[0]=col_array[1]=symbols[g1[i]];
KL_score[i] = get_KL (col_array, col_all, 2, rows);
KL_score_c += KL_score[i];
col_num++;
}
}
KL_score_c = KL_score_c/col_num;
for (i = 0; i < rows; i++)
{
arr_rows[i] = intersect_row(colcand, arr_c[dsItem(genes,0)], arr_c[i]);
arr_rows_b[i] = arr_rows[i];
}
/*we just get the largest 100 rows when we initial a bicluster because we believe that
* the 100 rows can characterize the structure of the bicluster
* btw, it can reduce the time complexity*/
if (rows > 100)
{
qsort(arr_rows_b, rows, sizeof *arr_rows, compare_int);
top = arr_rows_b[rows -100];
for (i = 0; i < rows; i++)
if (arr_rows[i] < top)
candidates[i] = FALSE;
}
/*calculate the condition low bound for current seed*/
int cutoff = floor (0.05*rows);
b->cond_low_bound = arr_rows_b[rows-cutoff];
while (*components < rows)
{
max_cnt = -1;
max_i = -1;
(*components)++;
cnt_all =0;
cnt_ave = 0;
/******************************************************/
/*add a function of controling the bicluster by pvalue*/
/******************************************************/
for (i=0; i< rows; i++)
{
if (!candidates[i]) continue;
if (po->IS_list && !sublist[i]) continue;
cnt = intersect_row(colcand,arr_c[dsItem(genes,0)],arr_c[i]);
cnt_all += cnt;
if (cnt < cand_threshold)
candidates[i] = FALSE;
if (cnt > max_cnt)
{
max_cnt = cnt;
max_i = i;
}
}
cnt_ave = cnt_all/row_all;
/*pvalue = get_pvalue (cnt_ave, max_cnt);*/
/* reconsider the genes with cnt=max_cnt when expand current bicluster base on the cwm-like significant of each row */
if (max_cnt>0)
{
KL_score_r = 0;
/*KL_score_max = 0;
for (i=0; i< rows; i++)
{
if (!candidates[i]) continue;
if (po->IS_list && !sublist[i]) continue;
cnt = intersect_row(colcand,arr_c[dsItem(genes,0)],arr_c[i]);
if (cnt == max_cnt)
{
sub_array = get_intersect_row(colcand,arr_c[dsItem(genes,0)],arr_c[i],cnt);
KL_score = get_KL (sub_array, arr_c[i], cnt, cols);
if (KL_score > KL_score_max)
{
max_cnt = cnt;
max_i = i;
}
}
}*/
sub_array = get_intersect_row(colcand,arr_c[dsItem(genes,0)],arr_c[max_i],max_cnt);
KL_score_r = get_KL (sub_array, arr_c[max_i], max_cnt, cols);
for (j=0;j<*components-1;j++)
KL_score_r += get_KL (sub_array, arr_c[dsItem(genes,j)], max_cnt, cols);
KL_score_r = KL_score_r/(*components);
for (j=0;j<cols;j++)
{
if (colcand[j] && (arr_c[max_i][j] != arr_c[dsItem(genes,0)][j]))
{
KL_score_c += (KL_score_c-KL_score[j])/(col_num-1);
col_num--;
}
}
pvalue = max_cnt/cnt_ave;
}
/*printf ("%d\t%.2f\t%.2f\t%.2f\n",max_cnt, KL_score_r, KL_score_c, pvalue);*/
if (po->IS_cond)
{
if (max_cnt < po->COL_WIDTH || max_i < 0 || max_cnt < b->cond_low_bound) break;
}
else
{
if (max_cnt < po->COL_WIDTH || max_i < 0) break;
}
/*printf ("%d\t%d\n",*components, max_cnt);*/
if (po->IS_area)
score = *components*max_cnt;
else
score = floor(100*(KL_score_r+KL_score_c));
/* score = floor(100*KL_score_r*pvalue);*/
/* score = MIN(*components, max_cnt);*/
if (score > b->score)
{
b->score = score;
b->significance = KL_score_r;
}
dsPush(genes, max_i);
dsPush(scores,score);
update_colcand(colcand,arr_c[dsItem(genes,0)], arr_c[max_i]);
candidates[max_i] = FALSE;
}
/*be sure to free a pointer when you finish using it*/
free(colcand);
}
static void block_init(Edge *e, Block *b,
struct dyStack *genes, struct dyStack *scores,
bool *candidates, const int cand_threshold,
int *components, struct dyStack *allincluster)
{
int i,score,top;
int cnt = 0;
int max_cnt, max_i;
int *arr_rows, *arr_rows_b;
AllocArray(arr_rows, rows);
AllocArray(arr_rows_b, rows);
bool *colcand;
AllocArray(colcand, cols);
for (i=0; i< cols; i++) colcand[i] = FALSE;
discrete *g1, *g2;
g1 = arr_c[dsItem(genes,0)];
g2 = arr_c[dsItem(genes,1)];
for (i=0; i< cols; i++)
if ((g1[i] == g2[i])&&(g1[i]!=0)) colcand[i] = TRUE;
for (i = 0; i < rows; i++)
{
arr_rows[i] = intersect_row(colcand, arr_c[dsItem(genes,0)], arr_c[i]);
arr_rows_b[i] = arr_rows[i];
}
if (rows > 100)
{
qsort(arr_rows_b, rows, sizeof *arr_rows, compare_int);
top = arr_rows_b[rows -100];
for (i = 0; i < rows; i++)
if (arr_rows[i] < top) candidates[i] = FALSE;
}
while (*components < rows)
{
max_cnt = -1;
max_i = -1;
(*components)++;
for (i=0; i< rows; i++)
{
if (!candidates[i]) continue;
cnt = intersect_row(colcand,arr_c[dsItem(genes,0)],arr_c[i]);
if (cnt < cand_threshold) candidates[i] = FALSE;
if (cnt > max_cnt)
{
max_cnt = cnt;
max_i = i;
}
}
if (max_cnt < po->COL_WIDTH || max_i < 0) break;
else
{
score = MIN(*components, max_cnt);
if (score > b->score) b->score = score;
dsPush(genes, max_i);
dsPush(scores,score);
update_colcand(colcand,arr_c[dsItem(genes,0)], arr_c[max_i]);
candidates[max_i] = FALSE;
}
}
free(colcand);
}