static void print_suggest_cutoff(size_t hist_distsize, size_t hist_covgsize,
uint64_t (*hists)[hist_covgsize],
FILE *fh)
{
size_t i, dist, median, nthresh_failed = 0;
size_t cutoffs[hist_distsize], sumcovgs[hist_distsize];
// Don't use dist[0] -- not informative
memset(cutoffs, 0, sizeof(cutoffs));
memset(sumcovgs, 0, sizeof(sumcovgs));
for(dist = 1; dist < hist_distsize; dist++)
{
int t = cleaning_pick_kmer_threshold(hists[dist], hist_covgsize,
NULL, NULL, NULL, NULL);
if(t < 0) { nthresh_failed++; t = 0; }
cutoffs[dist] = t;
for(i = 0; i < hist_covgsize; i++) sumcovgs[dist] += hists[dist][i];
}
// Print cutoffs
fprintf(fh, "sumcovgs=%zu", sumcovgs[1]);
for(i = 2; i < hist_distsize; i++) fprintf(fh, ",%zu", sumcovgs[i]);
fprintf(fh, "\ncutoffs=%zu", cutoffs[1]);
for(i = 2; i < hist_distsize; i++) fprintf(fh, ",%zu", cutoffs[i]);
fprintf(fh, "\n");
median = gca_median_size(cutoffs+1, hist_distsize-1);
fprintf(fh, "suggested_cutoff=%zu\n", median);
if(nthresh_failed)
warn("Threshold failed in %zu cases [default to 0]", nthresh_failed);
}
/**
* Get coverage threshold for removing unitigs
*
* @param visited should be at least db_graph.ht.capcity bits long and initialised
* to zero. On return, it will be 1 at each original kmer index
* @param covgs_csv_path
* @param lens_csv_path paths to files to write CSV histogram of unitigs
coverages and lengths BEFORE ANY CLEANING.
* If NULL these are ignored.
* @return threshold to clean or -1 on error
*/
int cleaning_get_threshold(size_t num_threads,
const char *covgs_csv_path,
const char *lens_csv_path,
uint8_t *visited,
const dBGraph *db_graph)
{
// Estimate optimum cleaning threshold
status("[cleaning] Calculating unitig stats with %zu threads...", num_threads);
status("[cleaning] Using kmer gamma method");
// Get kmer coverages and unitig lengths
UnitigCleaner cl;
unitig_cleaner_alloc(&cl, num_threads, 0, 0, NULL, db_graph);
supernodes_iterate(num_threads, visited, db_graph, unitig_get_covg, &cl);
// Get kmer coverage only (faster)
// KmerCleanerIterator kcls[nthreads];
// for(i = 0; i < nthreads; i++)
// kcls[i] = (KmerCleanerIterator){.threadid = i, .nthreads = nthreads, .cl = &cl};
// util_run_threads(kcls, nthreads, sizeof(kcls[0]), nthreads, kmer_get_covg);
// Wipe visited kmer memory
memset(visited, 0, roundup_bits2bytes(db_graph->ht.capacity));
if(covgs_csv_path != NULL) {
cleaning_write_covg_histogram(covgs_csv_path,
cl.kmer_covgs_init,
cl.unitig_covgs_init,
cl.covg_arrsize);
}
if(lens_csv_path != NULL) {
cleaning_write_len_histogram(lens_csv_path,
cl.len_hist_init,
cl.len_arrsize,
db_graph->kmer_size);
}
// set threshold using histogram and genome size
double alpha = 0, beta = 0, false_pos = 0, false_neg = 0;
int threshold_est = cleaning_pick_kmer_threshold(cl.kmer_covgs_init,
cl.covg_arrsize,
&alpha, &beta,
&false_pos, &false_neg);
if(threshold_est < 0)
warn("Cannot pick a cleaning threshold");
else {
status("[cleaning] alpha=%f, beta=%f FP=%f FN=%f",
alpha, beta, false_pos, false_neg);
status("[cleaning] Recommended unitig cleaning threshold: < %i",
threshold_est);
}
unitig_cleaner_dealloc(&cl);
return threshold_est;
}
/**
* Get coverage threshold for removing supernodes
*
* @param visited should be at least db_graph.ht.capcity bits long and initialised
* to zero. On return, it will be 1 at each original kmer index
* @param covgs_csv_path
* @param lens_csv_path paths to files to write CSV histogram of supernodes
coverages and lengths BEFORE ANY CLEANING.
* If NULL these are ignored.
* @return threshold to clean or -1 on error
*/
int cleaning_get_threshold(size_t num_threads,
const char *covgs_csv_path,
const char *lens_csv_path,
uint8_t *visited,
const dBGraph *db_graph)
{
// Estimate optimum cleaning threshold
status("[cleaning] Calculating supernode stats with %zu threads...", num_threads);
status("[cleaning] Using kmer gamma method");
// Get kmer coverages and supernode lengths
SupernodeCleaner cl;
supernode_cleaner_alloc(&cl, num_threads, 0, 0, NULL, db_graph);
supernodes_iterate(num_threads, visited, db_graph, supernode_get_covg, &cl);
// Get kmer coverage only (faster)
// KmerCleanerIterator kcls[nthreads];
// for(i = 0; i < nthreads; i++)
// kcls[i] = (KmerCleanerIterator){.threadid = i, .nthreads = nthreads, .cl = &cl};
// util_run_threads(kcls, nthreads, sizeof(kcls[0]), nthreads, kmer_get_covg);
// Wipe visited kmer memory
memset(visited, 0, roundup_bits2bytes(db_graph->ht.capacity));
if(covgs_csv_path != NULL) {
cleaning_write_covg_histogram(covgs_csv_path,
cl.covg_hist_init,
cl.mean_covg_hist_init,
cl.covg_arrsize);
}
if(lens_csv_path != NULL) {
cleaning_write_len_histogram(lens_csv_path,
cl.len_hist_init,
cl.len_arrsize,
db_graph->kmer_size);
}
// set threshold using histogram and genome size
int threshold_est = -1;
double fdr = 0.001, alpha = 0, beta = 0;
while(fdr < 1) {
threshold_est = cleaning_pick_kmer_threshold(cl.covg_hist_init,
cl.covg_arrsize,
fdr, &alpha, &beta);
if(threshold_est >= 0) break;
fdr *= 10;
}
if(threshold_est < 0)
warn("Cannot pick a cleaning threshold");
else
status("[cleaning] FDR set to %f [alpha=%f, beta=%f]", fdr, alpha, beta);
if(threshold_est >= 0) {
status("[cleaning] Recommended supernode cleaning threshold: < %i",
threshold_est);
}
supernode_cleaner_dealloc(&cl);
return threshold_est;
}
