// Merge temporary files, closes tmp files
void futil_merge_tmp_files(FILE **tmp_files, size_t num_files, FILE *fout)
{
#define TMP_BUF_SIZE (32 * ONE_MEGABYTE)
char *data = ctx_malloc(TMP_BUF_SIZE);
size_t i, len;
FILE *fh;
for(i = 0; i < num_files; i++)
{
fh = tmp_files[i];
if(fseek(fh, 0L, SEEK_SET) != 0) die("fseek error");
while((len = fread(data, 1, TMP_BUF_SIZE, fh)) > 0)
if(fwrite(data, 1, len, fout) != len)
die("write error [%s]", strerror(errno));
if(ferror(fh)) warn("fread error: %s", strerror(errno));
fclose(fh);
}
ctx_free(data);
#undef TMP_BUF_SIZE
}
// Returns sorted array of hkey_t from the hash table
hkey_t* hash_table_sorted(const HashTable *htable)
{
ctx_assert(sizeof(hkey_t) == sizeof(BinaryKmer*));
ctx_assert(sizeof(hkey_t) == sizeof(BkmerPtrHkeyUnion));
BkmerPtrHkeyUnion *kmers, *nxt, *end;
nxt = kmers = ctx_malloc(sizeof(BkmerPtrHkeyUnion) * htable->num_kmers);
end = kmers + htable->num_kmers;
HASH_ITERATE(htable, _fetch_kmer_union, htable, &nxt);
// Can sort ignoring that the top flag bit is set on all kmers
qsort(kmers, htable->num_kmers, sizeof(BinaryKmer*), binary_kmers_qcmp_ptrs);
for(nxt = kmers; nxt < end; nxt++) nxt->h = nxt->bptr - htable->table;
return (hkey_t*)kmers;
}
/**
* Calculate cleaning threshold for supernodes from a given distribution
* of supernode coverages
* @param covgs histogram of supernode coverages
*/
size_t cleaning_pick_supernode_threshold(const uint64_t *covgs, size_t len,
double seq_depth,
const dBGraph *db_graph)
{
ctx_assert(len > 5);
ctx_assert(db_graph->ht.num_kmers > 0);
size_t i, d1len = len-2, d2len = len-3, f1, f2;
double *tmp = ctx_malloc((d1len+d2len) * sizeof(double));
double *delta1 = tmp, *delta2 = tmp + d1len;
// Get sequencing depth from coverage
uint64_t covg_sum = 0, capacity = db_graph->ht.capacity * db_graph->num_of_cols;
for(i = 0; i < capacity; i++) covg_sum += db_graph->col_covgs[i];
double seq_depth_est = (double)covg_sum / db_graph->ht.num_kmers;
status("[cleaning] Kmer depth before cleaning supernodes: %.2f", seq_depth_est);
if(seq_depth <= 0) seq_depth = seq_depth_est;
else status("[cleaning] Using sequence depth argument: %f", seq_depth);
size_t fallback_thresh = (size_t)MAX2(1, (seq_depth+1)/2);
// +1 to ensure covgs is never 0
for(i = 0; i < d1len; i++) delta1[i] = (double)(covgs[i+1]+1) / (covgs[i+2]+1);
d1len = i;
d2len = d1len - 1;
if(d1len <= 2) {
status("[cleaning] (using fallback1)\n");
ctx_free(tmp);
return fallback_thresh;
}
// d2len is d1len-1
for(i = 0; i < d2len; i++) delta2[i] = delta1[i] / delta1[i+1];
for(f1 = 0; f1 < d1len && delta1[f1] >= 1; f1++);
for(f2 = 0; f2 < d2len && delta2[f2] > 1; f2++);
ctx_free(tmp);
if(f1 < d1len && f1 < (seq_depth*0.75))
{ status("[cleaning] (using f1)"); return f1+1; }
else if(f2 < d2len)
{ status("[cleaning] (using f2)"); return f2+1; }
else
{ status("[cleaning] (using fallback1)"); return fallback_thresh+1; }
}
void chrom_hash_load(char const*const* paths, size_t num_files,
ReadBuffer *chroms, ChromHash *genome)
{
size_t i;
seq_file_t **ref_files = ctx_malloc(num_files * sizeof(seq_file_t*));
for(i = 0; i < num_files; i++)
if((ref_files[i] = seq_open(paths[i])) == NULL)
die("Cannot read sequence file: %s", paths[i]);
chrom_hash_load2(ref_files, num_files, chroms, genome);
ctx_free(ref_files);
}
void hash_table_alloc(HashTable *ht, uint64_t req_capacity)
{
uint64_t num_of_buckets, capacity;
uint8_t bucket_size;
capacity = hash_table_cap(req_capacity, &num_of_buckets, &bucket_size);
uint_fast32_t hash_mask = (uint_fast32_t)(num_of_buckets - 1);
size_t mem = capacity * sizeof(BinaryKmer) +
num_of_buckets * sizeof(uint8_t[2]);
char num_bkts_str[100], bkt_size_str[100], cap_str[100], mem_str[100];
ulong_to_str(num_of_buckets, num_bkts_str);
ulong_to_str(bucket_size, bkt_size_str);
ulong_to_str(capacity, cap_str);
bytes_to_str(mem, 1, mem_str);
status("[hasht] Allocating table with %s entries, using %s", cap_str, mem_str);
status("[hasht] number of buckets: %s, bucket size: %s", num_bkts_str, bkt_size_str);
// calloc is required for bucket_data to set the first element of each bucket
// to the 0th pos
BinaryKmer *table = ctx_malloc(capacity * sizeof(BinaryKmer));
uint8_t (*const buckets)[2] = ctx_calloc(num_of_buckets, sizeof(uint8_t[2]));
size_t i;
for(i = 0; i < capacity; i++) table[i] = unset_bkmer;
HashTable data = {
.table = table,
.num_of_buckets = num_of_buckets,
.hash_mask = hash_mask,
.bucket_size = bucket_size,
.capacity = capacity,
.buckets = buckets,
.num_kmers = 0,
.collisions = {0},
.seed = rand()};
memcpy(ht, &data, sizeof(data));
}
void hash_table_dealloc(HashTable *hash_table)
{
ctx_free(hash_table->table);
ctx_free(hash_table->buckets);
}
// Merge temporary files, closes tmp files
void futil_merge_tmp_files(FILE **tmp_files, size_t num_files, FILE *fout)
{
#define TMP_BUF_SIZE (1<<25) /* 32MB */
char *data = ctx_malloc(TMP_BUF_SIZE);
size_t i, len;
FILE *tmp_file;
for(i = 0; i < num_files; i++)
{
tmp_file = tmp_files[i];
if(fseek(tmp_file, 0L, SEEK_SET) == -1) die("gzseek error");
while((len = fread(data, 1, TMP_BUF_SIZE, tmp_file)) > 0)
if(fwrite(data, 1, len, fout) != len)
die("write error [%s]", strerror(errno));
fclose(tmp_file);
}
ctx_free(data);
#undef TMP_BUF_SIZE
}
int ctx_thread(int argc, char **argv)
{
struct ReadThreadCmdArgs args;
read_thread_args_alloc(&args);
read_thread_args_parse(&args, argc, argv, longopts, false);
GraphFileReader *gfile = &args.gfile;
GPathFileBuffer *gpfiles = &args.gpfiles;
CorrectAlnInputBuffer *inputs = &args.inputs;
size_t i;
if(args.zero_link_counts && gpfiles->len == 0)
cmd_print_usage("-0,--zero-paths without -p,--paths <in.ctp> has no meaning");
// Check each path file only loads one colour
gpaths_only_for_colour(gpfiles->b, gpfiles->len, 0);
//
// Decide on memory
//
size_t bits_per_kmer, kmers_in_hash, graph_mem, total_mem;
size_t path_hash_mem, path_store_mem, path_mem;
bool sep_path_list = (!args.use_new_paths && gpfiles->len > 0);
bits_per_kmer = sizeof(BinaryKmer)*8 + sizeof(Edges)*8 + sizeof(GPath*)*8 +
2 * args.nthreads; // Have traversed
// false -> don't use mem_to_use to decide how many kmers to store in hash
// since we need some of that memory for storing paths
kmers_in_hash = cmd_get_kmers_in_hash(args.memargs.mem_to_use,
args.memargs.mem_to_use_set,
args.memargs.num_kmers,
args.memargs.num_kmers_set,
bits_per_kmer,
gfile->num_of_kmers,
gfile->num_of_kmers,
false, &graph_mem);
// Paths memory
size_t min_path_mem = 0;
gpath_reader_sum_mem(gpfiles->b, gpfiles->len, 1, true, true, &min_path_mem);
if(graph_mem + min_path_mem > args.memargs.mem_to_use) {
char buf[50];
die("Require at least %s memory", bytes_to_str(graph_mem+min_path_mem, 1, buf));
}
path_mem = args.memargs.mem_to_use - graph_mem;
size_t pentry_hash_mem = sizeof(GPEntry)/0.7;
size_t pentry_store_mem = sizeof(GPath) + 8 + // struct + sequence
1 + // in colour
sizeof(uint8_t) + // counts
sizeof(uint32_t); // kmer length
size_t max_paths = path_mem / (pentry_store_mem + pentry_hash_mem);
path_store_mem = max_paths * pentry_store_mem;
path_hash_mem = max_paths * pentry_hash_mem;
cmd_print_mem(path_hash_mem, "paths hash");
cmd_print_mem(path_store_mem, "paths store");
total_mem = graph_mem + path_mem;
cmd_check_mem_limit(args.memargs.mem_to_use, total_mem);
//
// Open output file
//
gzFile gzout = futil_gzopen_create(args.out_ctp_path, "w");
status("Creating paths file: %s", futil_outpath_str(args.out_ctp_path));
//
// Allocate memory
//
dBGraph db_graph;
size_t kmer_size = gfile->hdr.kmer_size;
db_graph_alloc(&db_graph, kmer_size, 1, 1, kmers_in_hash,
DBG_ALLOC_EDGES | DBG_ALLOC_NODE_IN_COL);
// Split path memory 2:1 between store and hash
// Create a path store that tracks path counts
gpath_store_alloc(&db_graph.gpstore,
db_graph.num_of_cols, db_graph.ht.capacity,
0, path_store_mem, true, sep_path_list);
// Create path hash table for fast lookup
gpath_hash_alloc(&db_graph.gphash, &db_graph.gpstore, path_hash_mem);
if(args.use_new_paths) {
status("Using paths as they are added (risky)");
} else {
status("Not using new paths as they are added (safe)");
}
//
// Start up workers to add paths to the graph
//
GenPathWorker *workers;
workers = gen_paths_workers_alloc(args.nthreads, &db_graph);
// Setup for loading graphs graph
LoadingStats gstats;
loading_stats_init(&gstats);
// Path statistics
LoadingStats *load_stats = gen_paths_get_stats(workers);
CorrectAlnStats *aln_stats = gen_paths_get_aln_stats(workers);
// Load contig hist distribution
for(i = 0; i < gpfiles->len; i++) {
gpath_reader_load_contig_hist(gpfiles->b[i].json,
gpfiles->b[i].fltr.path.b,
file_filter_fromcol(&gpfiles->b[i].fltr, 0),
&aln_stats->contig_histgrm);
}
GraphLoadingPrefs gprefs = {.db_graph = &db_graph,
.boolean_covgs = false,
.must_exist_in_graph = false,
.must_exist_in_edges = NULL,
.empty_colours = false}; // already loaded paths
// Load graph, print stats, close file
graph_load(gfile, gprefs, &gstats);
hash_table_print_stats_brief(&db_graph.ht);
graph_file_close(gfile);
// Load existing paths
for(i = 0; i < gpfiles->len; i++)
gpath_reader_load(&gpfiles->b[i], GPATH_DIE_MISSING_KMERS, &db_graph);
// zero link counts of already loaded links
if(args.zero_link_counts) {
status("Zeroing link counts for loaded links");
gpath_set_zero_nseen(&db_graph.gpstore.gpset);
}
if(!args.use_new_paths)
gpath_store_split_read_write(&db_graph.gpstore);
// Deal with a set of files at once
// Can have different numbers of inputs vs threads
size_t start, end;
for(start = 0; start < inputs->len; start += MAX_IO_THREADS)
{
end = MIN2(inputs->len, start+MAX_IO_THREADS);
generate_paths(inputs->b+start, end-start, workers, args.nthreads);
}
// Print memory statistics
gpath_hash_print_stats(&db_graph.gphash);
gpath_store_print_stats(&db_graph.gpstore);
correct_aln_dump_stats(aln_stats, load_stats,
args.dump_seq_sizes,
args.dump_frag_sizes,
db_graph.ht.num_kmers);
// Don't need GPathHash anymore
gpath_hash_dealloc(&db_graph.gphash);
cJSON **hdrs = ctx_malloc(gpfiles->len * sizeof(cJSON*));
for(i = 0; i < gpfiles->len; i++) hdrs[i] = gpfiles->b[i].json;
size_t output_threads = MIN2(args.nthreads, MAX_IO_THREADS);
// Generate a cJSON header for all inputs
cJSON *thread_hdr = cJSON_CreateObject();
cJSON *inputs_hdr = cJSON_CreateArray();
cJSON_AddItemToObject(thread_hdr, "inputs", inputs_hdr);
for(i = 0; i < inputs->len; i++)
cJSON_AddItemToArray(inputs_hdr, correct_aln_input_json_hdr(&inputs->b[i]));
// Write output file
gpath_save(gzout, args.out_ctp_path, output_threads, true,
"thread", thread_hdr, hdrs, gpfiles->len,
&aln_stats->contig_histgrm, 1,
&db_graph);
gzclose(gzout);
ctx_free(hdrs);
// Optionally run path checks for debugging
// gpath_checks_all_paths(&db_graph, args.nthreads);
// ins_gap, err_gap no longer allocated after this line
gen_paths_workers_dealloc(workers, args.nthreads);
// Close and free input files etc.
read_thread_args_dealloc(&args);
db_graph_dealloc(&db_graph);
return EXIT_SUCCESS;
}
int ctx_sort(int argc, char **argv)
{
const char *out_path = NULL;
struct MemArgs memargs = MEM_ARGS_INIT;
// Arg parsing
char cmd[100];
char shortopts[300];
cmd_long_opts_to_short(longopts, shortopts, sizeof(shortopts));
int c;
// silence error messages from getopt_long
// opterr = 0;
while((c = getopt_long_only(argc, argv, shortopts, longopts, NULL)) != -1) {
cmd_get_longopt_str(longopts, c, cmd, sizeof(cmd));
switch(c) {
case 0: /* flag set */ break;
case 'h': cmd_print_usage(NULL); break;
case 'f': cmd_check(!futil_get_force(), cmd); futil_set_force(true); break;
case 'm': cmd_mem_args_set_memory(&memargs, optarg); break;
case 'n': cmd_mem_args_set_nkmers(&memargs, optarg); break;
case 'o': cmd_check(!out_path, cmd); out_path = optarg; break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" sort -h` for help. Bad option: %s", argv[optind-1]);
default: die("Bad option: [%c]: %s", c, cmd);
}
}
if(optind+1 != argc)
cmd_print_usage("Require exactly one input graph file (.ctx)");
const char *ctx_path = argv[optind];
//
// Open Graph file
//
GraphFileReader gfile;
memset(&gfile, 0, sizeof(GraphFileReader));
graph_file_open2(&gfile, ctx_path, out_path ? "r" : "r+", true, 0);
if(!file_filter_is_direct(&gfile.fltr))
die("Cannot open graph file with a filter ('in.ctx:blah' syntax)");
size_t num_kmers, memory;
// Reading from a stream
if(gfile.num_of_kmers < 0) {
if(!memargs.num_kmers_set)
die("If reading from a stream, must give -n <num_kmers>");
num_kmers = memargs.num_kmers;
}
else num_kmers = gfile.num_of_kmers;
// Open output path (if given)
FILE *fout = out_path ? futil_fopen_create(out_path, "w") : NULL;
size_t i;
size_t ncols = gfile.hdr.num_of_cols;
size_t kmer_mem = sizeof(BinaryKmer) + (sizeof(Edges)+sizeof(Covg))*ncols;
memory = (sizeof(char*) + kmer_mem) * num_kmers;
char mem_str[50];
bytes_to_str(memory, 1, mem_str);
if(memory > memargs.mem_to_use)
die("Require at least %s memory", mem_str);
status("[memory] Total: %s", mem_str);
char *mem = ctx_malloc(kmer_mem * num_kmers);
char **kmers = ctx_malloc(num_kmers*sizeof(char*));
// Read in whole file
// if(graph_file_fseek(gfile, gfile.hdr_size, SEEK_SET) != 0) die("fseek failed");
size_t nkread = gfr_fread_bytes(&gfile, mem, num_kmers*kmer_mem);
if(nkread != num_kmers*kmer_mem)
die("Could only read %zu bytes [<%zu]", nkread, num_kmers*kmer_mem);
// check we are at the end of the file
char tmpc;
if(gfr_fread_bytes(&gfile, &tmpc, 1) != 0) {
die("More kmers in file than believed (kmers: %zu ncols: %zu).",
num_kmers, ncols);
}
status("Read %zu kmers with %zu colour%s", num_kmers,
ncols, util_plural_str(ncols));
for(i = 0; i < num_kmers; i++)
kmers[i] = mem + kmer_mem*i;
sort_block(kmers, num_kmers);
// Print
if(out_path != NULL) {
// saving to a different destination - write header
graph_write_header(fout, &gfile.hdr);
}
else {
// Directly manipulating gfile.fh here, using it to write later
// Not doing any more reading
if(fseek(gfile.fh, gfile.hdr_size, SEEK_SET) != 0) die("fseek failed");
fout = gfile.fh;
}
for(i = 0; i < num_kmers; i++)
if(fwrite(kmers[i], 1, kmer_mem, fout) != kmer_mem)
die("Cannot write to file");
if(out_path) fclose(fout);
graph_file_close(&gfile);
ctx_free(kmers);
ctx_free(mem);
return EXIT_SUCCESS;
}
int ctx_index(int argc, char **argv)
{
const char *out_path = NULL;
size_t block_size = 0, block_kmers = 0;
// Arg parsing
char cmd[100];
char shortopts[300];
cmd_long_opts_to_short(longopts, shortopts, sizeof(shortopts));
int c;
// silence error messages from getopt_long
// opterr = 0;
while((c = getopt_long_only(argc, argv, shortopts, longopts, NULL)) != -1) {
cmd_get_longopt_str(longopts, c, cmd, sizeof(cmd));
switch(c) {
case 0: /* flag set */ break;
case 'h': cmd_print_usage(NULL); break;
case 'o': cmd_check(!out_path, cmd); out_path = optarg; break;
case 'b':
cmd_check(!block_kmers, cmd);
block_kmers = cmd_size_nonzero(cmd, optarg);
break;
case 's':
cmd_check(!block_size, cmd);
block_size = cmd_size_nonzero(cmd, optarg);
break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" index -h` for help. Bad option: %s", argv[optind-1]);
default: abort();
}
}
if(optind+1 != argc)
cmd_print_usage("Require exactly one input graph file (.ctx)");
if(block_size && block_kmers)
cmd_print_usage("Cannot use --block-kmers and --block-size together");
const char *ctx_path = argv[optind];
//
// Open Graph file
//
GraphFileReader gfile;
memset(&gfile, 0, sizeof(GraphFileReader));
graph_file_open2(&gfile, ctx_path, "r+", true, 0);
if(!file_filter_is_direct(&gfile.fltr))
die("Cannot open graph file with a filter ('in.ctx:blah' syntax)");
// Open output file
FILE *fout = out_path ? futil_fopen_create(out_path, "w") : stdout;
// Start
size_t filencols = gfile.hdr.num_of_cols;
size_t kmer_size = gfile.hdr.kmer_size;
const char *path = file_filter_path(&gfile.fltr);
size_t ncols = file_filter_into_ncols(&gfile.fltr);
size_t kmer_mem = sizeof(BinaryKmer) + (sizeof(Edges)+sizeof(Covg))*filencols;
if(block_size) {
block_kmers = block_size / kmer_mem;
} else if(!block_size && !block_kmers) {
block_size = 4 * ONE_MEGABYTE;
block_kmers = block_size / kmer_mem;
}
// Update block-size
block_size = block_kmers * kmer_mem;
status("[index] block bytes: %zu kmers: %zu; kmer bytes: %zu, hdr: %zu",
block_size, block_kmers, kmer_mem, (size_t)gfile.hdr_size);
if(block_kmers == 0) die("Cannot set block_kmers to zero");
// Print header
fputs("#block_start\tnext_block\tfirst_kmer\tkmer_idx\tnext_kmer_idx\n", fout);
BinaryKmer bkmer = BINARY_KMER_ZERO_MACRO;
BinaryKmer prev_bkmer = BINARY_KMER_ZERO_MACRO;
Covg *covgs = ctx_malloc(ncols * sizeof(Covg));
Edges *edges = ctx_malloc(ncols * sizeof(Edges));
char bkmerstr[MAX_KMER_SIZE+1];
size_t rem_block = block_size - kmer_mem; // block after first kmer
char *tmp_mem = ctx_malloc(rem_block);
// Read in file, print index
size_t nblocks = 0;
size_t bl_bytes = 0, bl_kmers = 0;
size_t bl_byte_offset = gfile.hdr_size, bl_kmer_offset = 0;
while(1)
{
if(!graph_file_read(&gfile, &bkmer, covgs, edges)) {
status("Read kmer failed"); break; }
binary_kmer_to_str(bkmer, kmer_size, bkmerstr);
if(nblocks > 0 && !binary_kmer_less_than(prev_bkmer,bkmer))
die("File is not sorted: %s [%s]", bkmerstr, path);
// We've already read one kmer entry, read rest of block
bl_bytes = kmer_mem + gfr_fread_bytes(&gfile, tmp_mem, rem_block);
bl_kmers = 1 + bl_bytes / kmer_mem;
fprintf(fout, "%zu\t%zu\t%s\t%zu\t%zu\n",
bl_byte_offset, bl_byte_offset+bl_bytes, bkmerstr,
bl_kmer_offset, bl_kmer_offset+bl_kmers);
bl_byte_offset += bl_bytes;
bl_kmer_offset += bl_kmers;
nblocks++;
if(bl_kmers < block_kmers) {
status("last block %zu < %zu; %zu vs %zu",
bl_kmers, block_kmers, bl_bytes, block_size);
break;
}
prev_bkmer = bkmer;
}
ctx_free(covgs);
ctx_free(edges);
ctx_free(tmp_mem);
// done
char num_kmers_str[50], num_blocks_str[50];
char block_mem_str[50], block_kmers_str[50];
ulong_to_str(bl_kmer_offset, num_kmers_str);
ulong_to_str(nblocks, num_blocks_str);
bytes_to_str(block_size, 1, block_mem_str);
ulong_to_str(block_kmers, block_kmers_str);
status("Read %s kmers in %s block%s (block size %s / %s kmers)",
num_kmers_str, num_blocks_str, util_plural_str(nblocks),
block_mem_str, block_kmers_str);
if(fout != stdout) status("Saved to %s", out_path);
graph_file_close(&gfile);
fclose(fout);
return EXIT_SUCCESS;
}
// GenPathWorker stores four buffers of size n
#define gworker_seq_buf(n) (binary_seq_mem(n)*4)
static void _gen_paths_worker_alloc(GenPathWorker *wrkr, dBGraph *db_graph)
{
GenPathWorker tmp = {.db_graph = db_graph};
correct_aln_worker_alloc(&tmp.corrector, true, db_graph);
// Junction data
// only fw arrays are malloc'd, rv point to fw
tmp.junc_arrsize = INIT_BUFLEN;
tmp.pck_fw = ctx_calloc(2, gworker_seq_buf(tmp.junc_arrsize));
tmp.pck_rv = tmp.pck_fw + gworker_seq_buf(tmp.junc_arrsize);
tmp.pos_fw = ctx_malloc(tmp.junc_arrsize * sizeof(*tmp.pos_fw) * 2);
tmp.pos_rv = tmp.pos_fw + tmp.junc_arrsize;
tmp.num_fw = tmp.num_rv = 0;
memcpy(wrkr, &tmp, sizeof(GenPathWorker));
}
static void _gen_paths_worker_dealloc(GenPathWorker *wrkr)
{
correct_aln_worker_dealloc(&wrkr->corrector);
ctx_free(wrkr->pck_fw);
ctx_free(wrkr->pos_fw);
}
GenPathWorker* gen_paths_workers_alloc(size_t n, dBGraph *graph)
static BreakpointCaller* brkpt_callers_new(size_t num_callers,
gzFile gzout,
size_t min_ref_flank,
size_t max_ref_flank,
const KOGraph kograph,
const dBGraph *db_graph)
{
ctx_assert(num_callers > 0);
const size_t ncols = db_graph->num_of_cols;
BreakpointCaller *callers = ctx_malloc(num_callers * sizeof(BreakpointCaller));
pthread_mutex_t *out_lock = ctx_malloc(sizeof(pthread_mutex_t));
if(pthread_mutex_init(out_lock, NULL) != 0) die("mutex init failed");
size_t *callid = ctx_calloc(1, sizeof(size_t));
// Each colour in each caller can have a GraphCache path at once
PathRefRun *path_ref_runs = ctx_calloc(num_callers*MAX_REFRUNS_PER_CALLER(ncols),
sizeof(PathRefRun));
size_t i;
for(i = 0; i < num_callers; i++)
{
BreakpointCaller tmp = {.threadid = i,
.nthreads = num_callers,
.kograph = kograph,
.db_graph = db_graph,
.gzout = gzout,
.out_lock = out_lock,
.callid = callid,
.allele_refs = path_ref_runs,
.flank5p_refs = path_ref_runs+MAX_REFRUNS_PER_ORIENT(ncols),
.min_ref_nkmers = min_ref_flank,
.max_ref_nkmers = max_ref_flank};
memcpy(&callers[i], &tmp, sizeof(BreakpointCaller));
path_ref_runs += MAX_REFRUNS_PER_CALLER(ncols);
db_node_buf_alloc(&callers[i].allelebuf, 1024);
db_node_buf_alloc(&callers[i].flank5pbuf, 1024);
kmer_run_buf_alloc(&callers[i].koruns_5p, 128);
kmer_run_buf_alloc(&callers[i].koruns_5p_ended, 128);
kmer_run_buf_alloc(&callers[i].koruns_3p, 128);
kmer_run_buf_alloc(&callers[i].koruns_3p_ended, 128);
kmer_run_buf_alloc(&callers[i].allele_run_buf, 128);
kmer_run_buf_alloc(&callers[i].flank5p_run_buf, 128);
graph_crawler_alloc(&callers[i].crawlers[0], db_graph);
graph_crawler_alloc(&callers[i].crawlers[1], db_graph);
}
return callers;
}
static void brkpt_callers_destroy(BreakpointCaller *callers, size_t num_callers)
{
size_t i;
for(i = 0; i < num_callers; i++) {
db_node_buf_dealloc(&callers[i].allelebuf);
db_node_buf_dealloc(&callers[i].flank5pbuf);
kmer_run_buf_dealloc(&callers[i].koruns_5p);
kmer_run_buf_dealloc(&callers[i].koruns_5p_ended);
kmer_run_buf_dealloc(&callers[i].koruns_3p);
kmer_run_buf_dealloc(&callers[i].koruns_3p_ended);
kmer_run_buf_dealloc(&callers[i].allele_run_buf);
kmer_run_buf_dealloc(&callers[i].flank5p_run_buf);
graph_crawler_dealloc(&callers[i].crawlers[0]);
graph_crawler_dealloc(&callers[i].crawlers[1]);
}
pthread_mutex_destroy(callers[0].out_lock);
ctx_free(callers[0].out_lock);
ctx_free(callers[0].callid);
ctx_free(callers[0].allele_refs);
ctx_free(callers);
}
BubbleCaller* bubble_callers_new(size_t num_callers,
BubbleCallingPrefs prefs,
gzFile gzout,
const dBGraph *db_graph)
{
ctx_assert(num_callers > 0);
// Max usage is 4 * max_allele_len * cols
size_t i;
size_t max_path_len = MAX2(prefs.max_flank_len, prefs.max_allele_len);
BubbleCaller *callers = ctx_malloc(num_callers * sizeof(BubbleCaller));
pthread_mutex_t *out_lock = ctx_malloc(sizeof(pthread_mutex_t));
if(pthread_mutex_init(out_lock, NULL) != 0) die("mutex init failed");
size_t *num_bubbles_ptr = ctx_calloc(1, sizeof(size_t));
for(i = 0; i < num_callers; i++)
{
BubbleCaller tmp = {.threadid = i, .nthreads = num_callers,
.haploid_seen = ctx_calloc(1+prefs.num_haploid, sizeof(bool)),
.num_bubbles_ptr = num_bubbles_ptr,
.prefs = prefs,
.db_graph = db_graph, .gzout = gzout,
.out_lock = out_lock};
memcpy(&callers[i], &tmp, sizeof(BubbleCaller));
// First two buffers don't actually need to grow
db_node_buf_alloc(&callers[i].flank5p, prefs.max_flank_len);
db_node_buf_alloc(&callers[i].pathbuf, max_path_len);
graph_walker_alloc(&callers[i].wlk, db_graph);
rpt_walker_alloc(&callers[i].rptwlk, db_graph->ht.capacity, 22); // 4MB
graph_cache_alloc(&callers[i].cache, db_graph);
cache_stepptr_buf_alloc(&callers[i].spp_forward, 1024);
cache_stepptr_buf_alloc(&callers[i].spp_reverse, 1024);
strbuf_alloc(&callers[i].output_buf, 2048);
}
return callers;
}
void bubble_callers_destroy(BubbleCaller *callers, size_t num_callers)
{
ctx_assert(num_callers > 0);
size_t i;
for(i = 0; i < num_callers; i++)
{
ctx_free(callers[i].haploid_seen);
db_node_buf_dealloc(&callers[i].flank5p);
db_node_buf_dealloc(&callers[i].pathbuf);
rpt_walker_dealloc(&callers[i].rptwlk);
graph_walker_dealloc(&callers[i].wlk);
graph_cache_dealloc(&callers[i].cache);
cache_stepptr_buf_dealloc(&callers[i].spp_forward);
cache_stepptr_buf_dealloc(&callers[i].spp_reverse);
strbuf_dealloc(&callers[i].output_buf);
}
pthread_mutex_destroy(callers[0].out_lock);
ctx_free(callers[0].out_lock);
ctx_free(callers[0].num_bubbles_ptr);
ctx_free(callers);
}
int ctx_links(int argc, char **argv)
{
size_t limit = 0;
const char *link_out_path = NULL, *csv_out_path = NULL, *plot_out_path = NULL;
const char *thresh_path = NULL, *hist_path = NULL;
size_t hist_distsize = 0, hist_covgsize = 0;
size_t cutoff = 0;
bool clean = false;
// Arg parsing
char cmd[100];
char shortopts[300];
cmd_long_opts_to_short(longopts, shortopts, sizeof(shortopts));
int c;
while((c = getopt_long_only(argc, argv, shortopts, longopts, NULL)) != -1) {
cmd_get_longopt_str(longopts, c, cmd, sizeof(cmd));
switch(c) {
case 0: /* flag set */ break;
case 'h': cmd_print_usage(NULL); break;
case 'o': cmd_check(!link_out_path, cmd); link_out_path = optarg; break;
case 'f': cmd_check(!futil_get_force(), cmd); futil_set_force(true); break;
case 'l': cmd_check(!csv_out_path, cmd); csv_out_path = optarg; break;
case 'c': cmd_check(!cutoff, cmd); cutoff = cmd_size(cmd, optarg); clean = true; break;
case 'L': cmd_check(!limit, cmd); limit = cmd_size(cmd, optarg); break;
case 'P': cmd_check(!plot_out_path, cmd); plot_out_path = optarg; break;
case 'T': cmd_check(!thresh_path, cmd); thresh_path = optarg; break;
case 'H': cmd_check(!hist_path, cmd); hist_path = optarg; break;
case 'C': cmd_check(!hist_covgsize, cmd); hist_covgsize = cmd_size(cmd, optarg); break;
case 'D': cmd_check(!hist_distsize, cmd); hist_distsize = cmd_size(cmd, optarg); break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" links -h` for help. Bad option: %s", argv[optind-1]);
default: ctx_assert2(0, "shouldn't reach here: %c", c);
}
}
if(hist_distsize && !hist_path) cmd_print_usage("--max-dist without --covg-hist");
if(hist_covgsize && !hist_path) cmd_print_usage("--max-covg without --covg-hist");
// Defaults
if(!hist_distsize) hist_distsize = DEFAULT_MAX_DIST;
if(!hist_covgsize) hist_covgsize = DEFAULT_MAX_COVG;
if(optind + 1 != argc) cmd_print_usage("Wrong number of arguments");
const char *ctp_path = argv[optind];
bool list = (csv_out_path != NULL);
bool plot = (plot_out_path != NULL);
bool save = (link_out_path != NULL);
bool hist_covg = (thresh_path != NULL || hist_path != NULL);
size_t plot_kmer_idx = (limit == 0 ? 0 : limit - 1);
if(clean && !save)
cmd_print_usage("Need to give --out <out.ctp.gz> with --clean");
if(!save && !list && !plot && !hist_covg)
cmd_print_usage("Please specify one of --plot, --list or --clean");
if(link_out_path && hist_covg && strcmp(link_out_path,"-") == 0)
cmd_print_usage("Outputing both cleaning threshold (-T) and links (-o) to STDOUT!");
// Open input file
FILE *list_fh = NULL, *plot_fh = NULL, *link_tmp_fh = NULL;
FILE *thresh_fh = NULL, *hist_fh = NULL;
gzFile link_gz = NULL;
// Check file don't exist or that we can overwrite
// Will ignore if path is null
bool err = false;
err |= futil_check_outfile(csv_out_path);
err |= futil_check_outfile(plot_out_path);
err |= futil_check_outfile(link_out_path);
err |= futil_check_outfile(thresh_path);
err |= futil_check_outfile(hist_path);
if(err) die("Use -f,--force to overwrite files");
StrBuf link_tmp_path;
strbuf_alloc(&link_tmp_path, 1024);
GPathReader ctpin;
memset(&ctpin, 0, sizeof(ctpin));
gpath_reader_open(&ctpin, ctp_path);
size_t ncols = file_filter_into_ncols(&ctpin.fltr);
size_t kmer_size = gpath_reader_get_kmer_size(&ctpin);
cJSON *newhdr = cJSON_Duplicate(ctpin.json, 1);
if(ncols != 1) die("Can only clean a single colour at a time. Sorry.");
uint64_t (*hists)[hist_covgsize] = NULL;
if(hist_covg) {
hists = ctx_calloc(hist_distsize, sizeof(hists[0]));
}
if(hist_path && (hist_fh = futil_fopen_create(hist_path, "w")) == NULL)
die("Cannot open file: %s", hist_path);
if(thresh_path && (thresh_fh = futil_fopen_create(thresh_path, "w")) == NULL)
die("Cannot open file: %s", thresh_path);
if(limit)
status("Limiting to the first %zu kmers", limit);
if(clean)
{
timestamp();
message(" Cleaning coverage below %zu", cutoff);
message("\n");
}
if(save)
{
// Check we can find the fields we need
cJSON *links_json = json_hdr_get(newhdr, "paths", cJSON_Object, link_out_path);
cJSON *nkmers_json = json_hdr_get(links_json, "num_kmers_with_paths", cJSON_Number, link_out_path);
cJSON *nlinks_json = json_hdr_get(links_json, "num_paths", cJSON_Number, link_out_path);
cJSON *nbytes_json = json_hdr_get(links_json, "path_bytes", cJSON_Number, link_out_path);
if(!nkmers_json || !nlinks_json || !nbytes_json)
die("Cannot find required header entries");
// Create a random temporary file
link_tmp_fh = create_tmp_file(&link_tmp_path, link_out_path);
status("Saving output to: %s", link_out_path);
status("Temporary output: %s", link_tmp_path.b);
// Open output file
if((link_gz = futil_gzopen_create(link_out_path, "w")) == NULL)
die("Cannot open output link file: %s", link_out_path);
// Need to open output file first so we can get absolute path
// Update the header to include this command
json_hdr_add_curr_cmd(newhdr, link_out_path);
}
if(list)
{
status("Listing to %s", csv_out_path);
if((list_fh = futil_fopen_create(csv_out_path, "w")) == NULL)
die("Cannot open output CSV file %s", csv_out_path);
// Print csv header
fprintf(list_fh, "SeqLen,Covg\n");
}
if(plot)
{
status("Plotting kmer %zu to %s", plot_kmer_idx, plot_out_path);
if((plot_fh = futil_fopen_create(plot_out_path, "w")) == NULL)
die("Cannot open output .dot file %s", plot_out_path);
}
SizeBuffer countbuf, jposbuf;
size_buf_alloc(&countbuf, 16);
size_buf_alloc(&jposbuf, 1024);
StrBuf kmerbuf, juncsbuf, seqbuf, outbuf;
strbuf_alloc(&kmerbuf, 1024);
strbuf_alloc(&juncsbuf, 1024);
strbuf_alloc(&seqbuf, 1024);
strbuf_alloc(&outbuf, 1024);
bool link_fw;
size_t njuncs;
size_t knum, nlinks, num_links_exp = 0;
LinkTree ltree;
ltree_alloc(<ree, kmer_size);
LinkTreeStats tree_stats;
memset(&tree_stats, 0, sizeof(tree_stats));
size_t init_num_links = 0, num_links = 0;
for(knum = 0; !limit || knum < limit; knum++)
{
ltree_reset(<ree);
if(!gpath_reader_read_kmer(&ctpin, &kmerbuf, &num_links_exp)) break;
ctx_assert2(kmerbuf.end == kmer_size, "Kmer incorrect length %zu != %zu",
kmerbuf.end, kmer_size);
// status("kmer: %s", kmerbuf.b);
for(nlinks = 0;
gpath_reader_read_link(&ctpin, &link_fw, &njuncs,
&countbuf, &juncsbuf,
&seqbuf, &jposbuf);
nlinks++)
{
ltree_add(<ree, link_fw, countbuf.b[0], jposbuf.b,
juncsbuf.b, seqbuf.b);
}
if(nlinks != num_links_exp)
warn("Links count mismatch %zu != %zu", nlinks, num_links_exp);
if(hist_covg)
{
ltree_update_covg_hists(<ree, (uint64_t*)hists,
hist_distsize, hist_covgsize);
}
if(clean)
{
ltree_clean(<ree, cutoff);
}
// Accumulate statistics
ltree_get_stats(<ree, &tree_stats);
num_links = tree_stats.num_links - init_num_links;
init_num_links = tree_stats.num_links;
if(list)
{
ltree_write_list(<ree, &outbuf);
if(fwrite(outbuf.b, 1, outbuf.end, list_fh) != outbuf.end)
die("Cannot write CSV file to: %s", csv_out_path);
strbuf_reset(&outbuf);
}
if(save && num_links)
{
ltree_write_ctp(<ree, kmerbuf.b, num_links, &outbuf);
if(fwrite(outbuf.b, 1, outbuf.end, link_tmp_fh) != outbuf.end)
die("Cannot write ctp file to: %s", link_tmp_path.b);
strbuf_reset(&outbuf);
}
if(plot && knum == plot_kmer_idx)
{
status("Plotting tree...");
ltree_write_dot(<ree, &outbuf);
if(fwrite(outbuf.b, 1, outbuf.end, plot_fh) != outbuf.end)
die("Cannot write plot DOT file to: %s", plot_out_path);
strbuf_reset(&outbuf);
}
}
gpath_reader_close(&ctpin);
cJSON *links_json = json_hdr_get(newhdr, "paths", cJSON_Object, link_out_path);
cJSON *nkmers_json = json_hdr_get(links_json, "num_kmers_with_paths", cJSON_Number, link_out_path);
cJSON *nlinks_json = json_hdr_get(links_json, "num_paths", cJSON_Number, link_out_path);
cJSON *nbytes_json = json_hdr_get(links_json, "path_bytes", cJSON_Number, link_out_path);
status("Number of kmers with links %li -> %zu", nkmers_json->valueint, tree_stats.num_trees_with_links);
status("Number of links %li -> %zu", nlinks_json->valueint, tree_stats.num_links);
status("Number of bytes %li -> %zu", nbytes_json->valueint, tree_stats.num_link_bytes);
if(save)
{
// Update JSON
nkmers_json->valuedouble = nkmers_json->valueint = tree_stats.num_trees_with_links;
nlinks_json->valuedouble = nlinks_json->valueint = tree_stats.num_links;
nbytes_json->valuedouble = nbytes_json->valueint = tree_stats.num_link_bytes;
char *json_str = cJSON_Print(newhdr);
if(gzputs(link_gz, json_str) != (int)strlen(json_str))
die("Cannot write ctp file to: %s", link_out_path);
free(json_str);
gzputs(link_gz, "\n\n");
gzputs(link_gz, ctp_explanation_comment);
gzputs(link_gz, "\n");
fseek(link_tmp_fh, 0, SEEK_SET);
char *tmp = ctx_malloc(4*ONE_MEGABYTE);
size_t s;
while((s = fread(tmp, 1, 4*ONE_MEGABYTE, link_tmp_fh)) > 0) {
if(gzwrite(link_gz, tmp, s) != (int)s)
die("Cannot write to output: %s", link_out_path);
}
ctx_free(tmp);
gzclose(link_gz);
fclose(link_tmp_fh);
}
// Write histogram to file
if(hist_fh)
{
size_t i, j;
fprintf(hist_fh, " ");
for(j = 1; j < hist_covgsize; j++) fprintf(hist_fh, ",covg.%02zu", j);
fprintf(hist_fh, "\n");
for(i = 1; i < hist_distsize; i++) {
fprintf(hist_fh, "dist.%02zu", i);
for(j = 1; j < hist_covgsize; j++) {
fprintf(hist_fh, ",%"PRIu64, hists[i][j]);
}
fprintf(hist_fh, "\n");
}
}
if(thresh_fh)
{
// Use median of first five cutoffs
print_suggest_cutoff(6, hist_covgsize, hists, thresh_fh);
}
if(hist_fh && hist_fh != stdout) fclose(hist_fh);
if(list)
{
fclose(list_fh);
}
if(plot)
{
fclose(plot_fh);
}
ctx_free(hists);
cJSON_Delete(newhdr);
strbuf_dealloc(&link_tmp_path);
ltree_dealloc(<ree);
size_buf_dealloc(&countbuf);
size_buf_dealloc(&jposbuf);
strbuf_dealloc(&kmerbuf);
strbuf_dealloc(&juncsbuf);
strbuf_dealloc(&seqbuf);
strbuf_dealloc(&outbuf);
return EXIT_SUCCESS;
}
int ctx_bubbles(int argc, char **argv)
{
size_t nthreads = 0;
struct MemArgs memargs = MEM_ARGS_INIT;
const char *out_path = NULL;
size_t max_allele_len = 0, max_flank_len = 0;
bool remove_serial_bubbles = true;
// List of haploid colours
size_t *hapcols = NULL;
int nhapcols = 0;
char *hapcols_arg = NULL;
GPathReader tmp_gpfile;
GPathFileBuffer gpfiles;
gpfile_buf_alloc(&gpfiles, 8);
// Arg parsing
char cmd[100];
char shortopts[300];
cmd_long_opts_to_short(longopts, shortopts, sizeof(shortopts));
int c;
// silence error messages from getopt_long
// opterr = 0;
while((c = getopt_long_only(argc, argv, shortopts, longopts, NULL)) != -1) {
cmd_get_longopt_str(longopts, c, cmd, sizeof(cmd));
switch(c) {
case 0: /* flag set */ break;
case 'h': cmd_print_usage(NULL); break;
case 'o': cmd_check(!out_path, cmd); out_path = optarg; break;
case 'f': cmd_check(!futil_get_force(), cmd); futil_set_force(true); break;
case 'p':
memset(&tmp_gpfile, 0, sizeof(GPathReader));
gpath_reader_open(&tmp_gpfile, optarg);
gpfile_buf_push(&gpfiles, &tmp_gpfile, 1);
break;
case 't': cmd_check(!nthreads, cmd); nthreads = cmd_uint32_nonzero(cmd, optarg); break;
case 'm': cmd_mem_args_set_memory(&memargs, optarg); break;
case 'n': cmd_mem_args_set_nkmers(&memargs, optarg); break;
case 'H': cmd_check(!hapcols_arg, cmd); hapcols_arg = optarg; break;
case 'A': cmd_check(!max_allele_len, cmd); max_allele_len = cmd_uint32_nonzero(cmd, optarg); break;
case 'F': cmd_check(!max_flank_len, cmd); max_flank_len = cmd_uint32_nonzero(cmd, optarg); break;
case 'S': cmd_check(remove_serial_bubbles,cmd); remove_serial_bubbles = false; break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" "SUBCMD" -h` for help. Bad option: %s", argv[optind-1]);
default: abort();
}
}
// Defaults for unset values
if(out_path == NULL) out_path = "-";
if(nthreads == 0) nthreads = DEFAULT_NTHREADS;
if(max_allele_len == 0) max_allele_len = DEFAULT_MAX_ALLELE;
if(max_flank_len == 0) max_flank_len = DEFAULT_MAX_FLANK;
if(optind >= argc) cmd_print_usage("Require input graph files (.ctx)");
//
// Open graph files
//
const size_t num_gfiles = argc - optind;
char **graph_paths = argv + optind;
ctx_assert(num_gfiles > 0);
GraphFileReader *gfiles = ctx_calloc(num_gfiles, sizeof(GraphFileReader));
size_t i, ncols, ctx_max_kmers = 0, ctx_sum_kmers = 0;
ncols = graph_files_open(graph_paths, gfiles, num_gfiles,
&ctx_max_kmers, &ctx_sum_kmers);
// Check graph + paths are compatible
graphs_gpaths_compatible(gfiles, num_gfiles, gpfiles.b, gpfiles.len, -1);
//
// Check haploid colours are valid
//
if(hapcols_arg != NULL) {
if((nhapcols = range_get_num(hapcols_arg, ncols)) < 0)
die("Invalid haploid colour list: %s", hapcols_arg);
hapcols = ctx_calloc(nhapcols, sizeof(hapcols[0]));
if(range_parse_array(hapcols_arg, hapcols, ncols) < 0)
die("Invalid haploid colour list: %s", hapcols_arg);
}
//
// Decide on memory
//
size_t bits_per_kmer, kmers_in_hash, graph_mem, path_mem, thread_mem;
char thread_mem_str[100];
// edges(1bytes) + kmer_paths(8bytes) + in_colour(1bit/col) +
// visitedfw/rv(2bits/thread)
bits_per_kmer = sizeof(BinaryKmer)*8 + sizeof(Edges)*8 +
(gpfiles.len > 0 ? sizeof(GPath*)*8 : 0) +
ncols + 2*nthreads;
kmers_in_hash = cmd_get_kmers_in_hash(memargs.mem_to_use,
memargs.mem_to_use_set,
memargs.num_kmers,
memargs.num_kmers_set,
bits_per_kmer,
ctx_max_kmers, ctx_sum_kmers,
false, &graph_mem);
// Thread memory
thread_mem = roundup_bits2bytes(kmers_in_hash) * 2;
bytes_to_str(thread_mem * nthreads, 1, thread_mem_str);
status("[memory] (of which threads: %zu x %zu = %s)\n",
nthreads, thread_mem, thread_mem_str);
// Paths memory
size_t rem_mem = memargs.mem_to_use - MIN2(memargs.mem_to_use, graph_mem+thread_mem);
path_mem = gpath_reader_mem_req(gpfiles.b, gpfiles.len, ncols, rem_mem, false,
kmers_in_hash, false);
// Shift path store memory from graphs->paths
graph_mem -= sizeof(GPath*)*kmers_in_hash;
path_mem += sizeof(GPath*)*kmers_in_hash;
cmd_print_mem(path_mem, "paths");
size_t total_mem = graph_mem + thread_mem + path_mem;
cmd_check_mem_limit(memargs.mem_to_use, total_mem);
//
// Open output file
//
gzFile gzout = futil_gzopen_create(out_path, "w");
// Allocate memory
dBGraph db_graph;
db_graph_alloc(&db_graph, gfiles[0].hdr.kmer_size, ncols, 1, kmers_in_hash,
DBG_ALLOC_EDGES | DBG_ALLOC_NODE_IN_COL);
// Paths
gpath_reader_alloc_gpstore(gpfiles.b, gpfiles.len, path_mem, false, &db_graph);
//
// Load graphs
//
GraphLoadingPrefs gprefs = graph_loading_prefs(&db_graph);
gprefs.empty_colours = true;
for(i = 0; i < num_gfiles; i++) {
graph_load(&gfiles[i], gprefs, NULL);
graph_file_close(&gfiles[i]);
gprefs.empty_colours = false;
}
ctx_free(gfiles);
hash_table_print_stats(&db_graph.ht);
// Load link files
for(i = 0; i < gpfiles.len; i++)
gpath_reader_load(&gpfiles.b[i], GPATH_DIE_MISSING_KMERS, &db_graph);
// Create array of cJSON** from input files
cJSON **hdrs = ctx_malloc(gpfiles.len * sizeof(cJSON*));
for(i = 0; i < gpfiles.len; i++) hdrs[i] = gpfiles.b[i].json;
// Now call variants
BubbleCallingPrefs call_prefs = {.max_allele_len = max_allele_len,
.max_flank_len = max_flank_len,
.haploid_cols = hapcols,
.nhaploid_cols = nhapcols,
.remove_serial_bubbles = remove_serial_bubbles};
invoke_bubble_caller(nthreads, &call_prefs,
gzout, out_path,
hdrs, gpfiles.len,
&db_graph);
status(" saved to: %s\n", out_path);
gzclose(gzout);
ctx_free(hdrs);
// Close input link files
for(i = 0; i < gpfiles.len; i++)
gpath_reader_close(&gpfiles.b[i]);
gpfile_buf_dealloc(&gpfiles);
ctx_free(hapcols);
db_graph_dealloc(&db_graph);
return EXIT_SUCCESS;
}
int ctx_vcfcov(int argc, char **argv)
{
struct MemArgs memargs = MEM_ARGS_INIT;
const char *out_path = NULL, *out_type = NULL;
uint32_t max_allele_len = 0, max_gt_vars = 0;
char *ref_path = NULL;
bool low_mem = false;
// Arg parsing
char cmd[100];
char shortopts[300];
cmd_long_opts_to_short(longopts, shortopts, sizeof(shortopts));
int c;
size_t i;
// silence error messages from getopt_long
// opterr = 0;
while((c = getopt_long_only(argc, argv, shortopts, longopts, NULL)) != -1) {
cmd_get_longopt_str(longopts, c, cmd, sizeof(cmd));
switch(c) {
case 0: /* flag set */ break;
case 'h': cmd_print_usage(NULL); break;
case 'o': cmd_check(!out_path, cmd); out_path = optarg; break;
case 'O': cmd_check(!out_type, cmd); out_type = optarg; break;
case 'f': cmd_check(!futil_get_force(), cmd); futil_set_force(true); break;
case 'm': cmd_mem_args_set_memory(&memargs, optarg); break;
case 'n': cmd_mem_args_set_nkmers(&memargs, optarg); break;
case 'r': cmd_check(!ref_path, cmd); ref_path = optarg; break;
case 'L': cmd_check(!max_allele_len,cmd); max_allele_len = cmd_uint32(cmd,optarg); break;
case 'N': cmd_check(!max_gt_vars,cmd); max_gt_vars = cmd_uint32(cmd,optarg); break;
case 'M': cmd_check(!low_mem, cmd); low_mem = true; break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" "SUBCMD" -h` for help. Bad option: %s", argv[optind-1]);
default: abort();
}
}
// Defaults for unset values
if(out_path == NULL) out_path = "-";
if(ref_path == NULL) cmd_print_usage("Require a reference (-r,--ref <ref.fa>)");
if(optind+2 > argc) cmd_print_usage("Require VCF and graph files");
if(!max_allele_len) max_allele_len = DEFAULT_MAX_ALLELE_LEN;
if(!max_gt_vars) max_gt_vars = DEFAULT_MAX_GT_VARS;
status("[vcfcov] max allele length: %u; max number of variants: %u",
max_allele_len, max_gt_vars);
// open ref
// index fasta with: samtools faidx ref.fa
faidx_t *fai = fai_load(ref_path);
if(fai == NULL) die("Cannot load ref index: %s / %s.fai", ref_path, ref_path);
// Open input VCF file
const char *vcf_path = argv[optind++];
htsFile *vcffh = hts_open(vcf_path, "r");
if(vcffh == NULL) die("Cannot open VCF file: %s", vcf_path);
bcf_hdr_t *vcfhdr = bcf_hdr_read(vcffh);
if(vcfhdr == NULL) die("Cannot read VCF header: %s", vcf_path);
// Test we can close and reopen files
if(low_mem) {
if((vcffh = hts_open(vcf_path, "r")) == NULL)
die("Cannot re-open VCF file: %s", vcf_path);
if((vcfhdr = bcf_hdr_read(vcffh)) == NULL)
die("Cannot re-read VCF header: %s", vcf_path);
}
//
// Open graph files
//
const size_t num_gfiles = argc - optind;
char **graph_paths = argv + optind;
ctx_assert(num_gfiles > 0);
GraphFileReader *gfiles = ctx_calloc(num_gfiles, sizeof(GraphFileReader));
size_t ncols, ctx_max_kmers = 0, ctx_sum_kmers = 0;
ncols = graph_files_open(graph_paths, gfiles, num_gfiles,
&ctx_max_kmers, &ctx_sum_kmers);
// Check graph + paths are compatible
graphs_gpaths_compatible(gfiles, num_gfiles, NULL, 0, -1);
//
// Decide on memory
//
size_t bits_per_kmer, kmers_in_hash, graph_mem;
bits_per_kmer = sizeof(BinaryKmer)*8 + sizeof(Covg)*8 * ncols;
kmers_in_hash = cmd_get_kmers_in_hash(memargs.mem_to_use,
memargs.mem_to_use_set,
memargs.num_kmers,
memargs.num_kmers_set,
bits_per_kmer,
low_mem ? -1 : (int64_t)ctx_max_kmers,
ctx_sum_kmers,
true, &graph_mem);
cmd_check_mem_limit(memargs.mem_to_use, graph_mem);
//
// Open output file
//
// v=>vcf, z=>compressed vcf, b=>bcf, bu=>uncompressed bcf
int mode = vcf_misc_get_outtype(out_type, out_path);
futil_create_output(out_path);
htsFile *outfh = hts_open(out_path, modes_htslib[mode]);
status("[vcfcov] Output format: %s", hsmodes_htslib[mode]);
// Allocate memory
dBGraph db_graph;
db_graph_alloc(&db_graph, gfiles[0].hdr.kmer_size, ncols, 1, kmers_in_hash,
DBG_ALLOC_COVGS);
//
// Set up tag names
//
// *R => ref, *A => alt
sprintf(kcov_ref_tag, "K%zuR", db_graph.kmer_size); // mean coverage
sprintf(kcov_alt_tag, "K%zuA", db_graph.kmer_size);
// #SAMPLE=<ID=...,K29KCOV=...,K29NK=...,K29RLK>
// - K29_kcov is empirical kmer coverage
// - K29_nkmers is the number of kmers in the sample
// - mean_read_length is the mean read length in bases
char sample_kcov_tag[20], sample_nk_tag[20], sample_rlk_tag[20];
sprintf(sample_kcov_tag, "K%zu_kcov", db_graph.kmer_size); // mean coverage
sprintf(sample_nk_tag, "K%zu_nkmers", db_graph.kmer_size);
sprintf(sample_rlk_tag, "mean_read_length");
//
// Load kmers if we are using --low-mem
//
VcfCovStats st;
memset(&st, 0, sizeof(st));
VcfCovPrefs prefs = {.kcov_ref_tag = kcov_ref_tag,
.kcov_alt_tag = kcov_alt_tag,
.max_allele_len = max_allele_len,
.max_gt_vars = max_gt_vars,
.load_kmers_only = false};
if(low_mem)
{
status("[vcfcov] Loading kmers from VCF+ref");
prefs.load_kmers_only = true;
vcfcov_file(vcffh, vcfhdr, NULL, NULL, vcf_path, fai,
NULL, &prefs, &st, &db_graph);
// Close files
hts_close(vcffh);
bcf_hdr_destroy(vcfhdr);
// Re-open files
if((vcffh = hts_open(vcf_path, "r")) == NULL)
die("Cannot re-open VCF file: %s", vcf_path);
if((vcfhdr = bcf_hdr_read(vcffh)) == NULL)
die("Cannot re-read VCF header: %s", vcf_path);
prefs.load_kmers_only = false;
}
//
// Load graphs
//
GraphLoadingStats gstats;
memset(&gstats, 0, sizeof(gstats));
GraphLoadingPrefs gprefs = graph_loading_prefs(&db_graph);
gprefs.must_exist_in_graph = low_mem;
for(i = 0; i < num_gfiles; i++) {
graph_load(&gfiles[i], gprefs, &gstats);
graph_file_close(&gfiles[i]);
}
ctx_free(gfiles);
hash_table_print_stats(&db_graph.ht);
//
// Set up VCF header / graph matchup
//
size_t *samplehdrids = ctx_malloc(db_graph.num_of_cols * sizeof(size_t));
// Add samples to vcf header
bcf_hdr_t *outhdr = bcf_hdr_dup(vcfhdr);
bcf_hrec_t *hrec;
int sid;
char hdrstr[200];
for(i = 0; i < db_graph.num_of_cols; i++) {
char *sname = db_graph.ginfo[i].sample_name.b;
if((sid = bcf_hdr_id2int(outhdr, BCF_DT_SAMPLE, sname)) < 0) {
bcf_hdr_add_sample(outhdr, sname);
sid = bcf_hdr_id2int(outhdr, BCF_DT_SAMPLE, sname);
}
samplehdrids[i] = sid;
// Add SAMPLE field
hrec = bcf_hdr_get_hrec(outhdr, BCF_HL_STR, "ID", sname, "SAMPLE");
if(hrec == NULL) {
sprintf(hdrstr, "##SAMPLE=<ID=%s,%s=%"PRIu64",%s=%"PRIu64",%s=%zu>", sname,
sample_kcov_tag,
gstats.nkmers[i] ? gstats.sumcov[i] / gstats.nkmers[i] : 0,
sample_nk_tag, gstats.nkmers[i],
sample_rlk_tag, (size_t)db_graph.ginfo[i].mean_read_length);
bcf_hdr_append(outhdr, hdrstr);
}
else {
// mean kcovg
sprintf(hdrstr, "%"PRIu64, gstats.sumcov[i] / gstats.nkmers[i]);
vcf_misc_add_update_hrec(hrec, sample_kcov_tag, hdrstr);
// num kmers
sprintf(hdrstr, "%"PRIu64, gstats.nkmers[i]);
vcf_misc_add_update_hrec(hrec, sample_nk_tag, hdrstr);
// mean read length in kmers
sprintf(hdrstr, "%zu", (size_t)db_graph.ginfo[i].mean_read_length);
vcf_misc_add_update_hrec(hrec, sample_rlk_tag, hdrstr);
}
status("[vcfcov] Colour %zu: %s [VCF column %zu]", i, sname, samplehdrids[i]);
}
// Add genotype format fields
// One field per alternative allele
sprintf(hdrstr, "##FORMAT=<ID=%s,Number=A,Type=Integer,"
"Description=\"Coverage on ref (k=%zu): sum(kmer_covs) / exp_num_kmers\">\n",
kcov_ref_tag, db_graph.kmer_size);
bcf_hdr_append(outhdr, hdrstr);
sprintf(hdrstr, "##FORMAT=<ID=%s,Number=A,Type=Integer,"
"Description=\"Coverage on alt (k=%zu): sum(kmer_covs) / exp_num_kmers\">\n",
kcov_alt_tag, db_graph.kmer_size);
bcf_hdr_append(outhdr, hdrstr);
bcf_hdr_set_version(outhdr, "VCFv4.2");
// Add command string to header
vcf_misc_hdr_add_cmd(outhdr, cmd_get_cmdline(), cmd_get_cwd());
if(bcf_hdr_write(outfh, outhdr) != 0)
die("Cannot write header to: %s", futil_outpath_str(out_path));
status("[vcfcov] Reading %s and adding coverage", vcf_path);
// Reset stats and get coverage
memset(&st, 0, sizeof(st));
vcfcov_file(vcffh, vcfhdr, outfh, outhdr, vcf_path, fai,
samplehdrids, &prefs, &st, &db_graph);
// Print statistics
char ns0[50], ns1[50];
status("[vcfcov] Read %s VCF lines", ulong_to_str(st.nvcf_lines, ns0));
status("[vcfcov] Read %s ALTs", ulong_to_str(st.nalts_read, ns0));
status("[vcfcov] Used %s kmers", ulong_to_str(st.ngt_kmers, ns0));
status("[vcfcov] ALTs used: %s / %s (%.2f%%)",
ulong_to_str(st.nalts_loaded, ns0), ulong_to_str(st.nalts_read, ns1),
st.nalts_read ? (100.0*st.nalts_loaded) / st.nalts_read : 0.0);
status("[vcfcov] ALTs too long (>%ubp): %s / %s (%.2f%%)", max_allele_len,
ulong_to_str(st.nalts_too_long, ns0), ulong_to_str(st.nalts_read, ns1),
st.nalts_read ? (100.0*st.nalts_too_long) / st.nalts_read : 0.0);
status("[vcfcov] ALTs too dense (>%u within %zubp): %s / %s (%.2f%%)",
max_gt_vars, db_graph.kmer_size,
ulong_to_str(st.nalts_no_covg, ns0), ulong_to_str(st.nalts_read, ns1),
st.nalts_read ? (100.0*st.nalts_no_covg) / st.nalts_read : 0.0);
status("[vcfcov] ALTs printed with coverage: %s / %s (%.2f%%)",
ulong_to_str(st.nalts_with_covg, ns0), ulong_to_str(st.nalts_read, ns1),
st.nalts_read ? (100.0*st.nalts_with_covg) / st.nalts_read : 0.0);
status("[vcfcov] Saved to: %s\n", out_path);
ctx_free(samplehdrids);
graph_loading_stats_destroy(&gstats);
bcf_hdr_destroy(vcfhdr);
bcf_hdr_destroy(outhdr);
hts_close(vcffh);
hts_close(outfh);
fai_destroy(fai);
db_graph_dealloc(&db_graph);
return EXIT_SUCCESS;
}
