// Return max colours we load from, or 0 if no colours
uint32_t file_filter_from_ncols_calc(const FileFilter *fltr)
{
uint32_t i, max = 0;
for(i = 0; i < file_filter_num(fltr); i++)
max = MAX2(max, file_filter_fromcol(fltr,i)+1);
return max;
}
// Returns true if each colour is loaded directly into the same colour
// 0->0, ... N->N where N is filencols
bool file_filter_is_direct(const FileFilter *fltr)
{
size_t i = 0;
if(file_filter_num(fltr) != fltr->filencols) return false;
while(i < file_filter_num(fltr) &&
file_filter_fromcol(fltr, i) == file_filter_intocol(fltr,i)) i++;
return (i == file_filter_num(fltr));
}
// Print file filter description
void file_filter_status(const FileFilter *fltr)
{
size_t i;
pthread_mutex_lock(&ctx_biglock);
timestamp();
message("[FileFilter] Loading file %s [%u colour%s]", file_filter_path(fltr),
fltr->filencols, util_plural_str(fltr->filencols));
if(!file_filter_is_direct(fltr))
{
message(" with filter: %u->%u", file_filter_fromcol(fltr, 0),
file_filter_intocol(fltr, 0));
for(i = 1; i < file_filter_num(fltr); i++)
message(",%u->%u", file_filter_fromcol(fltr,i), file_filter_intocol(fltr,i));
}
message("\n");
pthread_mutex_unlock(&ctx_biglock);
}
// Read a kmer from the file
// returns true on success, false otherwise
// prints warnings if dirty kmers in file
// be sure to zero covgs, edges before reading in
bool graph_file_read(const GraphFileReader *file,
BinaryKmer *bkmer, Covg *covgs, Edges *edges)
{
// status("Header colours: %u", file->hdr.num_of_cols);
Covg kmercovgs[file->hdr.num_of_cols];
Edges kmeredges[file->hdr.num_of_cols];
size_t i, from, into;
const FileFilter *fltr = &file->fltr;
if(!graph_file_read_kmer(file->fh, &file->hdr, fltr->path.b,
bkmer, kmercovgs, kmeredges)) return false;
for(i = 0; i < file_filter_num(fltr); i++) {
from = file_filter_fromcol(fltr, i);
into = file_filter_intocol(fltr, i);
covgs[into] = SAFE_ADD_COVG(covgs[into], kmercovgs[from]);
edges[into] |= kmeredges[from];
}
return true;
}
int ctx_clean(int argc, char **argv)
{
size_t nthreads = 0, use_ncols = 0;
struct MemArgs memargs = MEM_ARGS_INIT;
const char *out_ctx_path = NULL;
bool tip_cleaning = false, supernode_cleaning = false;
size_t min_keep_tip = 0;
Covg threshold = 0, fallback_thresh = 0;
const char *len_before_path = NULL, *len_after_path = NULL;
const char *covg_before_path = NULL, *covg_after_path = NULL;
// 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(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 'o':
if(out_ctx_path != NULL) cmd_print_usage(NULL);
out_ctx_path = optarg;
break;
case 'm': cmd_mem_args_set_memory(&memargs, optarg); break;
case 'n': cmd_mem_args_set_nkmers(&memargs, optarg); break;
case 'N': use_ncols = cmd_uint32_nonzero(cmd, optarg); break;
case 't': cmd_check(!nthreads, cmd); nthreads = cmd_uint32_nonzero(cmd, optarg); break;
case 'T':
cmd_check(!tip_cleaning, cmd);
min_keep_tip = cmd_uint32_nonzero(cmd, optarg);
tip_cleaning = true;
break;
case 'S':
cmd_check(!supernode_cleaning, cmd);
if(optarg != NULL) threshold = cmd_uint32_nonzero(cmd, optarg);
supernode_cleaning = true;
break;
case 'B': cmd_check(!fallback_thresh, cmd); fallback_thresh = cmd_uint32_nonzero(cmd, optarg); break;
case 'l': cmd_check(!len_before_path, cmd); len_before_path = optarg; break;
case 'L': cmd_check(!len_after_path, cmd); len_after_path = optarg; break;
case 'c': cmd_check(!covg_before_path, cmd); covg_before_path = optarg; break;
case 'C': cmd_check(!covg_after_path, cmd); covg_after_path = optarg; break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" clean -h` for help. Bad option: %s", argv[optind-1]);
default: abort();
}
}
if(nthreads == 0) nthreads = DEFAULT_NTHREADS;
if(optind >= argc) cmd_print_usage("Please give input graph files");
// Default behaviour
if(!tip_cleaning && !supernode_cleaning) {
if(out_ctx_path != NULL)
supernode_cleaning = tip_cleaning = true; // do both
else
warn("No cleaning being done: you did not specify --out <out.ctx>");
}
bool doing_cleaning = (supernode_cleaning || tip_cleaning);
if(doing_cleaning && out_ctx_path == NULL) {
cmd_print_usage("Please specify --out <out.ctx> for cleaned graph");
}
if(!doing_cleaning && (covg_after_path || len_after_path)) {
cmd_print_usage("You gave --len-after <out> / --covg-after <out> without "
"any cleaning (set -s, --supernodes or -t, --tips)");
}
if(doing_cleaning && strcmp(out_ctx_path,"-") != 0 &&
!futil_get_force() && futil_file_exists(out_ctx_path))
{
cmd_print_usage("Output file already exists: %s", out_ctx_path);
}
if(fallback_thresh && !supernode_cleaning)
cmd_print_usage("-B, --fallback <T> without --supernodes");
// Use remaining args as graph files
char **gfile_paths = argv + optind;
size_t i, j, num_gfiles = (size_t)(argc - optind);
// Open graph files
GraphFileReader *gfiles = ctx_calloc(num_gfiles, sizeof(GraphFileReader));
size_t ncols, ctx_max_kmers = 0, ctx_sum_kmers = 0;
ncols = graph_files_open(gfile_paths, gfiles, num_gfiles,
&ctx_max_kmers, &ctx_sum_kmers);
size_t kmer_size = gfiles[0].hdr.kmer_size;
// default to one colour for now
if(use_ncols == 0) use_ncols = 1;
// Flatten if we don't have to remember colours / output a graph
if(!doing_cleaning)
{
ncols = use_ncols = 1;
for(i = 0; i < num_gfiles; i++)
file_filter_flatten(&gfiles[i].fltr, 0);
}
if(ncols < use_ncols) {
warn("I only need %zu colour%s ('--ncols %zu' ignored)",
ncols, util_plural_str(ncols), use_ncols);
use_ncols = ncols;
}
char max_kmers_str[100];
ulong_to_str(ctx_max_kmers, max_kmers_str);
status("%zu input graph%s, max kmers: %s, using %zu colours",
num_gfiles, util_plural_str(num_gfiles), max_kmers_str, use_ncols);
// If no arguments given we default to removing tips < 2*kmer_size
if(tip_cleaning && min_keep_tip == 0)
min_keep_tip = 2 * kmer_size;
// Warn if any graph files already cleaned
size_t fromcol, intocol;
ErrorCleaning *cleaning;
for(i = 0; i < num_gfiles; i++) {
for(j = 0; j < file_filter_num(&gfiles[i].fltr); j++) {
fromcol = file_filter_fromcol(&gfiles[i].fltr, j);
cleaning = &gfiles[i].hdr.ginfo[fromcol].cleaning;
if(cleaning->cleaned_snodes && supernode_cleaning) {
warn("%s:%zu already has supernode cleaning with threshold: <%zu",
file_filter_path(&gfiles[i].fltr), fromcol,
(size_t)cleaning->clean_snodes_thresh);
}
if(cleaning->cleaned_tips && tip_cleaning) {
warn("%s:%zu already has had tip cleaned",
file_filter_path(&gfiles[i].fltr), fromcol);
}
}
}
// Print steps
size_t step = 0;
status("Actions:\n");
if(covg_before_path != NULL)
status("%zu. Saving kmer coverage distribution to: %s", step++, covg_before_path);
if(len_before_path != NULL)
status("%zu. Saving supernode length distribution to: %s", step++, len_before_path);
if(tip_cleaning)
status("%zu. Cleaning tips shorter than %zu nodes", step++, min_keep_tip);
if(supernode_cleaning && threshold > 0)
status("%zu. Cleaning supernodes with coverage < %u", step++, threshold);
if(supernode_cleaning && threshold <= 0)
status("%zu. Cleaning supernodes with auto-detected threshold", step++);
if(covg_after_path != NULL)
status("%zu. Saving kmer coverage distribution to: %s", step++, covg_after_path);
if(len_after_path != NULL)
status("%zu. Saving supernode length distribution to: %s", step++, len_after_path);
//
// Decide memory usage
//
bool all_colours_loaded = (ncols <= use_ncols);
bool use_mem_limit = (memargs.mem_to_use_set && num_gfiles > 1) || !ctx_max_kmers;
size_t kmers_in_hash, bits_per_kmer, graph_mem;
size_t per_kmer_per_col_bits = (sizeof(BinaryKmer)+sizeof(Covg)+sizeof(Edges)) * 8;
size_t pop_edges_per_kmer_bits = (all_colours_loaded ? 0 : sizeof(Edges) * 8);
bits_per_kmer = per_kmer_per_col_bits * use_ncols + pop_edges_per_kmer_bits;
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,
use_mem_limit, &graph_mem);
// Maximise the number of colours we load to fill the mem
size_t max_usencols = (memargs.mem_to_use*8 - pop_edges_per_kmer_bits * kmers_in_hash) /
(per_kmer_per_col_bits * kmers_in_hash);
use_ncols = MIN2(max_usencols, ncols);
cmd_check_mem_limit(memargs.mem_to_use, graph_mem);
//
// Check output files are writable
//
futil_create_output(out_ctx_path);
// Does nothing if arg is NULL
futil_create_output(covg_before_path);
futil_create_output(covg_after_path);
futil_create_output(len_before_path);
futil_create_output(len_after_path);
// Create db_graph
// Load as many colours as possible
// Use an extra set of edge to take intersections
dBGraph db_graph;
db_graph_alloc(&db_graph, gfiles[0].hdr.kmer_size, use_ncols, use_ncols,
kmers_in_hash, DBG_ALLOC_COVGS);
// Edges is a special case
size_t num_edges = db_graph.ht.capacity * (use_ncols + !all_colours_loaded);
db_graph.col_edges = ctx_calloc(num_edges, sizeof(Edges));
// Load graph into a single colour
LoadingStats stats = LOAD_STATS_INIT_MACRO;
GraphLoadingPrefs gprefs = {.db_graph = &db_graph,
.boolean_covgs = false,
.must_exist_in_graph = false,
.must_exist_in_edges = NULL,
.empty_colours = false};
// Construct cleaned graph header
GraphFileHeader outhdr;
memset(&outhdr, 0, sizeof(GraphFileHeader));
outhdr.version = CTX_GRAPH_FILEFORMAT;
outhdr.kmer_size = db_graph.kmer_size;
outhdr.num_of_cols = ncols;
outhdr.num_of_bitfields = (db_graph.kmer_size*2+63)/64;
graph_header_alloc(&outhdr, ncols);
// Merge info into header
size_t gcol = 0;
for(i = 0; i < num_gfiles; i++) {
for(j = 0; j < file_filter_num(&gfiles[i].fltr); j++, gcol++) {
fromcol = file_filter_fromcol(&gfiles[i].fltr, j);
intocol = file_filter_intocol(&gfiles[i].fltr, j);
graph_info_merge(&outhdr.ginfo[intocol], &gfiles[i].hdr.ginfo[fromcol]);
}
}
if(ncols > use_ncols) {
graph_files_load_flat(gfiles, num_gfiles, gprefs, &stats);
} else {
for(i = 0; i < num_gfiles; i++)
graph_load(&gfiles[i], gprefs, &stats);
}
char num_kmers_str[100];
ulong_to_str(db_graph.ht.num_kmers, num_kmers_str);
status("Total kmers loaded: %s\n", num_kmers_str);
size_t initial_nkmers = db_graph.ht.num_kmers;
hash_table_print_stats(&db_graph.ht);
uint8_t *visited = ctx_calloc(roundup_bits2bytes(db_graph.ht.capacity), 1);
uint8_t *keep = ctx_calloc(roundup_bits2bytes(db_graph.ht.capacity), 1);
if((supernode_cleaning && threshold <= 0) || covg_before_path || len_before_path)
{
// Get coverage distribution and estimate cleaning threshold
int est_threshold = cleaning_get_threshold(nthreads,
covg_before_path,
len_before_path,
visited, &db_graph);
if(est_threshold < 0) status("Cannot find recommended cleaning threshold");
else status("Recommended cleaning threshold is: %i", est_threshold);
// Use estimated threshold if threshold not set
if(threshold <= 0) {
if(fallback_thresh > 0 && est_threshold < (int)fallback_thresh) {
status("Using fallback threshold: %i", fallback_thresh);
threshold = fallback_thresh;
}
else if(est_threshold >= 0) threshold = est_threshold;
}
}
// Die if we failed to find suitable cleaning threshold
if(supernode_cleaning && threshold <= 0)
die("Need cleaning threshold (--supernodes=<D> or --fallback <D>)");
if(doing_cleaning) {
// Clean graph of tips (if min_keep_tip > 0) and supernodes (if threshold > 0)
clean_graph(nthreads, threshold, min_keep_tip,
covg_after_path, len_after_path,
visited, keep, &db_graph);
}
ctx_free(visited);
ctx_free(keep);
if(doing_cleaning)
{
// Output graph file
Edges *intersect_edges = NULL;
bool kmers_loaded = true;
size_t col, thresh;
// Set output header ginfo cleaned
for(col = 0; col < ncols; col++)
{
cleaning = &outhdr.ginfo[col].cleaning;
cleaning->cleaned_snodes |= supernode_cleaning;
cleaning->cleaned_tips |= tip_cleaning;
// if(tip_cleaning) {
// strbuf_append_str(&outhdr.ginfo[col].sample_name, ".tipclean");
// }
if(supernode_cleaning) {
thresh = cleaning->clean_snodes_thresh;
thresh = cleaning->cleaned_snodes ? MAX2(thresh, (uint32_t)threshold)
: (uint32_t)threshold;
cleaning->clean_snodes_thresh = thresh;
// char name_append[200];
// sprintf(name_append, ".supclean%zu", thresh);
// strbuf_append_str(&outhdr.ginfo[col].sample_name, name_append);
}
}
if(!all_colours_loaded)
{
// We haven't loaded all the colours
// intersect_edges are edges to mask with
// resets graph edges
intersect_edges = db_graph.col_edges;
db_graph.col_edges += db_graph.ht.capacity;
}
// Print stats on removed kmers
size_t removed_nkmers = initial_nkmers - db_graph.ht.num_kmers;
double removed_pct = (100.0 * removed_nkmers) / initial_nkmers;
char removed_str[100], init_str[100];
ulong_to_str(removed_nkmers, removed_str);
ulong_to_str(initial_nkmers, init_str);
status("Removed %s of %s (%.2f%%) kmers", removed_str, init_str, removed_pct);
graph_files_merge(out_ctx_path, gfiles, num_gfiles,
kmers_loaded, all_colours_loaded,
intersect_edges, &outhdr, &db_graph);
// Swap back
if(!all_colours_loaded)
db_graph.col_edges = intersect_edges;
}
ctx_check(db_graph.ht.num_kmers == hash_table_count_kmers(&db_graph.ht));
graph_header_dealloc(&outhdr);
for(i = 0; i < num_gfiles; i++) graph_file_close(&gfiles[i]);
ctx_free(gfiles);
db_graph_dealloc(&db_graph);
return EXIT_SUCCESS;
}
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_clean(int argc, char **argv)
{
size_t nthreads = 0, use_ncols = 0;
struct MemArgs memargs = MEM_ARGS_INIT;
const char *out_ctx_path = NULL;
int min_keep_tip = -1, unitig_min = -1; // <0 => default, 0 => noclean
uint32_t fallback_thresh = 0;
const char *len_before_path = NULL, *len_after_path = NULL;
const char *covg_before_path = NULL, *covg_after_path = NULL;
// 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(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 'o':
if(out_ctx_path != NULL) cmd_print_usage(NULL);
out_ctx_path = optarg;
break;
case 'm': cmd_mem_args_set_memory(&memargs, optarg); break;
case 'n': cmd_mem_args_set_nkmers(&memargs, optarg); break;
case 'N': use_ncols = cmd_uint32_nonzero(cmd, optarg); break;
case 't': cmd_check(!nthreads, cmd); nthreads = cmd_uint32_nonzero(cmd, optarg); break;
case 'T':
cmd_check(min_keep_tip<0, cmd);
min_keep_tip = (optarg != NULL ? (int)cmd_uint32(cmd, optarg) : -1);
break;
case 'S':
case 'U':
cmd_check(unitig_min<0, cmd);
unitig_min = (optarg != NULL ? cmd_uint32(cmd, optarg) : -1);
break;
case 'B': cmd_check(!fallback_thresh, cmd); fallback_thresh = cmd_uint32_nonzero(cmd, optarg); break;
case 'l': cmd_check(!len_before_path, cmd); len_before_path = optarg; break;
case 'L': cmd_check(!len_after_path, cmd); len_after_path = optarg; break;
case 'c': cmd_check(!covg_before_path, cmd); covg_before_path = optarg; break;
case 'C': cmd_check(!covg_after_path, cmd); covg_after_path = optarg; break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
// cmd_print_usage(NULL);
die("`"CMD" clean -h` for help. Bad option: %s", argv[optind-1]);
default: abort();
}
}
if(nthreads == 0) nthreads = DEFAULT_NTHREADS;
if(optind >= argc) cmd_print_usage("Please give input graph files");
bool unitig_cleaning = (unitig_min != 0);
bool tip_cleaning = (min_keep_tip != 0);
bool doing_cleaning = (unitig_cleaning || tip_cleaning);
// If you ever want to estimate cleaning threshold without outputting
// a graph, change this to a warning
if(doing_cleaning && out_ctx_path == NULL) {
cmd_print_usage("Please specify --out <out.ctx> for cleaned graph");
// warn("No cleaning being done: you did not specify --out <out.ctx>");
}
if(!doing_cleaning && (covg_after_path || len_after_path)) {
warn("You gave --len-after <out> / --covg-after <out> without "
"any cleaning (set -U, --unitigs or -t, --tips)");
}
if(doing_cleaning && strcmp(out_ctx_path,"-") != 0 &&
!futil_get_force() && futil_file_exists(out_ctx_path))
{
cmd_print_usage("Output file already exists: %s", out_ctx_path);
}
if(fallback_thresh && !unitig_cleaning)
warn("-B, --fallback <T> without --unitigs");
// Use remaining args as graph files
char **gfile_paths = argv + optind;
size_t i, j, num_gfiles = (size_t)(argc - optind);
// Open graph files
GraphFileReader *gfiles = ctx_calloc(num_gfiles, sizeof(GraphFileReader));
size_t col, ncols, ctx_max_kmers = 0, ctx_sum_kmers = 0;
ncols = graph_files_open(gfile_paths, gfiles, num_gfiles,
&ctx_max_kmers, &ctx_sum_kmers);
size_t kmer_size = gfiles[0].hdr.kmer_size;
// default to one colour for now
if(use_ncols == 0) use_ncols = 1;
// Flatten if we don't have to remember colours / output a graph
if(out_ctx_path == NULL)
{
ncols = use_ncols = 1;
for(i = 0; i < num_gfiles; i++)
file_filter_flatten(&gfiles[i].fltr, 0);
}
if(ncols < use_ncols) {
warn("I only need %zu colour%s ('--ncols %zu' ignored)",
ncols, util_plural_str(ncols), use_ncols);
use_ncols = ncols;
}
char max_kmers_str[100];
ulong_to_str(ctx_max_kmers, max_kmers_str);
status("%zu input graph%s, max kmers: %s, using %zu colours",
num_gfiles, util_plural_str(num_gfiles), max_kmers_str, use_ncols);
// If no arguments given we default to removing tips < 2*kmer_size
if(min_keep_tip < 0)
min_keep_tip = 2 * kmer_size;
// Warn if any graph files already cleaned
size_t fromcol;
ErrorCleaning *cleaning;
for(i = 0; i < num_gfiles; i++) {
for(j = 0; j < file_filter_num(&gfiles[i].fltr); j++) {
fromcol = file_filter_fromcol(&gfiles[i].fltr, j);
cleaning = &gfiles[i].hdr.ginfo[fromcol].cleaning;
if(cleaning->cleaned_snodes && unitig_cleaning) {
warn("%s:%zu already has unitig cleaning with threshold: <%zu",
file_filter_path(&gfiles[i].fltr), fromcol,
(size_t)cleaning->clean_snodes_thresh);
}
if(cleaning->cleaned_tips && tip_cleaning) {
warn("%s:%zu already has had tip cleaned",
file_filter_path(&gfiles[i].fltr), fromcol);
}
}
}
// Print steps
size_t step = 0;
status("Actions:\n");
if(covg_before_path != NULL)
status("%zu. Saving kmer coverage distribution to: %s", step++, covg_before_path);
if(len_before_path != NULL)
status("%zu. Saving unitig length distribution to: %s", step++, len_before_path);
if(min_keep_tip > 0)
status("%zu. Cleaning tips shorter than %i nodes", step++, min_keep_tip);
if(unitig_min > 0)
status("%zu. Cleaning unitigs with coverage < %i", step++, unitig_min);
if(unitig_min < 0)
status("%zu. Cleaning unitigs with auto-detected threshold", step++);
if(covg_after_path != NULL)
status("%zu. Saving kmer coverage distribution to: %s", step++, covg_after_path);
if(len_after_path != NULL)
status("%zu. Saving unitig length distribution to: %s", step++, len_after_path);
//
// Decide memory usage
//
bool all_colours_loaded = (ncols <= use_ncols);
bool use_mem_limit = (memargs.mem_to_use_set && num_gfiles > 1) || !ctx_max_kmers;
size_t kmers_in_hash, bits_per_kmer, graph_mem;
size_t per_col_bits = (sizeof(Covg)+sizeof(Edges)) * 8;
size_t extra_edge_bits = (all_colours_loaded ? 0 : sizeof(Edges) * 8);
bits_per_kmer = sizeof(BinaryKmer)*8 +
per_col_bits * use_ncols +
extra_edge_bits;
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,
use_mem_limit, &graph_mem);
// Maximise the number of colours we load to fill the mem
size_t max_usencols = (memargs.mem_to_use*8 -
sizeof(BinaryKmer)*8*kmers_in_hash +
extra_edge_bits*kmers_in_hash) /
(per_col_bits*kmers_in_hash);
use_ncols = MIN2(max_usencols, ncols);
cmd_check_mem_limit(memargs.mem_to_use, graph_mem);
//
// Check output files are writable
//
futil_create_output(out_ctx_path);
// Does nothing if arg is NULL
futil_create_output(covg_before_path);
futil_create_output(covg_after_path);
futil_create_output(len_before_path);
futil_create_output(len_after_path);
// Create db_graph
// Load as many colours as possible
// Use an extra set of edge to take intersections
dBGraph db_graph;
db_graph_alloc(&db_graph, gfiles[0].hdr.kmer_size, use_ncols, use_ncols,
kmers_in_hash, DBG_ALLOC_EDGES | DBG_ALLOC_COVGS);
// Extra edges required to hold union of kept edges
Edges *edges_union = NULL;
if(use_ncols < ncols)
edges_union = ctx_calloc(db_graph.ht.capacity, sizeof(Edges));
// Load graph into a single colour
GraphLoadingPrefs gprefs = graph_loading_prefs(&db_graph);
// Construct cleaned graph header
GraphFileHeader outhdr;
memset(&outhdr, 0, sizeof(GraphFileHeader));
for(i = 0; i < num_gfiles; i++)
graph_file_merge_header(&outhdr, &gfiles[i]);
if(ncols > use_ncols)
{
db_graph.num_of_cols = db_graph.num_edge_cols = 1;
SWAP(edges_union, db_graph.col_edges);
graphs_load_files_flat(gfiles, num_gfiles, gprefs, NULL);
SWAP(edges_union, db_graph.col_edges);
db_graph.num_of_cols = db_graph.num_edge_cols = use_ncols;
}
else {
for(i = 0; i < num_gfiles; i++)
graph_load(&gfiles[i], gprefs, NULL);
}
char num_kmers_str[100];
ulong_to_str(db_graph.ht.num_kmers, num_kmers_str);
status("Total kmers loaded: %s\n", num_kmers_str);
size_t initial_nkmers = db_graph.ht.num_kmers;
hash_table_print_stats(&db_graph.ht);
uint8_t *visited = ctx_calloc(roundup_bits2bytes(db_graph.ht.capacity), 1);
uint8_t *keep = ctx_calloc(roundup_bits2bytes(db_graph.ht.capacity), 1);
// Always estimate cleaning threshold
// if(unitig_min <= 0 || covg_before_path || len_before_path)
// {
// Get coverage distribution and estimate cleaning threshold
int est_min_covg = cleaning_get_threshold(nthreads,
covg_before_path,
len_before_path,
visited, &db_graph);
if(est_min_covg < 0) status("Cannot find recommended cleaning threshold");
else status("Recommended cleaning threshold is: %i", est_min_covg);
// Use estimated threshold if threshold not set
if(unitig_min < 0) {
if(fallback_thresh > 0 && est_min_covg < (int)fallback_thresh) {
status("Using fallback threshold: %i", fallback_thresh);
unitig_min = fallback_thresh;
}
else if(est_min_covg >= 0) unitig_min = est_min_covg;
}
// }
// Die if we failed to find suitable cleaning threshold
if(unitig_min < 0)
die("Need cleaning threshold (--unitigs=<D> or --fallback <D>)");
// Cleaning parameters should now be set (>0) or turned off (==0)
ctx_assert(unitig_min >= 0);
ctx_assert(min_keep_tip >= 0);
if(unitig_min || min_keep_tip)
{
// Clean graph of tips (if min_keep_tip > 0) and unitigs (if threshold > 0)
clean_graph(nthreads, unitig_min, min_keep_tip,
covg_after_path, len_after_path,
visited, keep, &db_graph);
}
ctx_free(visited);
ctx_free(keep);
if(out_ctx_path != NULL)
{
// Set output header ginfo cleaned
for(col = 0; col < ncols; col++)
{
cleaning = &outhdr.ginfo[col].cleaning;
cleaning->cleaned_snodes |= unitig_cleaning;
cleaning->cleaned_tips |= tip_cleaning;
// if(tip_cleaning) {
// strbuf_append_str(&outhdr.ginfo[col].sample_name, ".tipclean");
// }
if(unitig_cleaning) {
size_t thresh = cleaning->clean_snodes_thresh;
thresh = cleaning->cleaned_snodes ? MAX2(thresh, (uint32_t)unitig_min)
: (uint32_t)unitig_min;
cleaning->clean_snodes_thresh = thresh;
// char name_append[200];
// sprintf(name_append, ".supclean%zu", thresh);
// strbuf_append_str(&outhdr.ginfo[col].sample_name, name_append);
}
}
// Print stats on removed kmers
size_t removed_nkmers = initial_nkmers - db_graph.ht.num_kmers;
double removed_pct = (100.0 * removed_nkmers) / initial_nkmers;
char removed_str[100], init_str[100];
ulong_to_str(removed_nkmers, removed_str);
ulong_to_str(initial_nkmers, init_str);
status("Removed %s of %s (%.2f%%) kmers", removed_str, init_str, removed_pct);
// kmers_loaded=true
graph_writer_merge(out_ctx_path, gfiles, num_gfiles,
true, all_colours_loaded,
edges_union, &outhdr, &db_graph);
}
ctx_check(db_graph.ht.num_kmers == hash_table_count_kmers(&db_graph.ht));
// TODO: report kmer coverage for each sample
graph_header_dealloc(&outhdr);
for(i = 0; i < num_gfiles; i++) graph_file_close(&gfiles[i]);
ctx_free(gfiles);
ctx_free(edges_union);
db_graph_dealloc(&db_graph);
return EXIT_SUCCESS;
}
// If there is no 'into' filter (e.g. 0,1:in.ctx 0,1 is 'into filter'),
// load into `into_offset..into_offset+N-1`
void file_filter_set_cols(FileFilter *fltr, size_t filencols, size_t into_offset)
{
size_t i;
const char *path_start_c, *path_end_c;
file_filter_deconstruct_path(fltr->input.b, &path_start_c, &path_end_c);
// This is a hack to get non-const pointer
size_t offset_start = path_start_c - fltr->input.b;
size_t offset_end = path_end_c - fltr->input.b;
char *path_start = fltr->input.b + offset_start;
char *path_end = fltr->input.b + offset_end;
char *from_fltr = (*path_end == ':' ? path_end+1 : NULL);
char *into_fltr = (path_start > fltr->input.b ? fltr->input.b : NULL);
size_t ncols;
if(from_fltr) {
int s = range_get_num(from_fltr, filencols-1);
if(s < 0)
die("Invalid filter path: %s (file cols: %zu)", fltr->input.b, filencols);
ncols = s;
} else {
ncols = filencols;
}
if(into_fltr) {
*(path_start-1) = '\0';
int s = range_get_num(into_fltr, SIZE_MAX);
*(path_start-1) = ':';
if(s < 0 || (s != 1 && (size_t)s != ncols))
die("Invalid filter path: %s (s:%i ncols:%zu)", fltr->input.b, s, ncols);
}
fltr->filencols = filencols;
filter_buf_capacity(&fltr->filter, ncols);
file_filter_num(fltr) = ncols;
size_t *tmp = ctx_calloc(ncols, sizeof(size_t));
if(from_fltr)
{
if(range_parse_array(from_fltr, tmp, filencols-1) == -1)
die("Invalid filter path: %s", fltr->input.b);
for(i = 0; i < ncols; i++)
file_filter_fromcol(fltr, i) = tmp[i];
}
else {
for(i = 0; i < ncols; i++)
file_filter_fromcol(fltr, i) = i;
}
if(into_fltr)
{
*(path_start-1) = '\0';
int s = range_parse_array_fill(into_fltr, tmp, SIZE_MAX, ncols);
*(path_start-1) = ':';
if(s < 0 || (size_t)s != ncols)
die("Invalid filter path: %s (s:%i ncols:%zu)", fltr->input.b, s, ncols);
for(i = 0; i < ncols; i++)
file_filter_intocol(fltr, i) = tmp[i];
}
else {
for(i = 0; i < ncols; i++)
file_filter_intocol(fltr, i) = into_offset + i;
}
ctx_free(tmp);
file_filter_update(fltr);
// for(i = 0; i < file_filter_num(fltr); i++)
// status("%u -> %u", file_filter_fromcol(fltr,i), file_filter_intocol(fltr,i));
}
int ctx_contigs(int argc, char **argv)
{
size_t nthreads = 0;
struct MemArgs memargs = MEM_ARGS_INIT;
const char *out_path = NULL;
size_t i, contig_limit = 0, colour = 0;
bool cmd_reseed = false, cmd_no_reseed = false; // -r, -R
const char *conf_table_path = NULL; // save confidence table to here
bool use_missing_info_check = true, seed_with_unused_paths = false;
double min_step_confid = -1.0, min_cumul_confid = -1.0; // < 0 => no min
// Read length and expected depth for calculating confidences
size_t genome_size = 0;
seq_file_t *tmp_seed_file = NULL;
SeqFilePtrBuffer seed_buf;
seq_file_ptr_buf_alloc(&seed_buf, 16);
GPathReader tmp_gpfile;
GPathFileBuffer gpfiles;
gpfile_buf_alloc(&gpfiles, 8);
// Arg parsing
char cmd[100], 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 'o': cmd_check(!out_path,cmd); out_path = optarg; 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 'p':
memset(&tmp_gpfile, 0, sizeof(GPathReader));
gpath_reader_open(&tmp_gpfile, optarg);
gpfile_buf_push(&gpfiles, &tmp_gpfile, 1);
break;
case '1':
case 's': // --seed <in.fa>
if((tmp_seed_file = seq_open(optarg)) == NULL)
die("Cannot read --seed file: %s", optarg);
seq_file_ptr_buf_add(&seed_buf, tmp_seed_file);
break;
case 'r': cmd_check(!cmd_reseed,cmd); cmd_reseed = true; break;
case 'R': cmd_check(!cmd_no_reseed,cmd); cmd_no_reseed = true; break;
case 'N':
cmd_check(!contig_limit,cmd);
contig_limit = cmd_uint32_nonzero(cmd, optarg);
break;
case 'c': cmd_check(!colour,cmd); colour = cmd_uint32(cmd, optarg); break;
case 'G': cmd_check(!genome_size,cmd); genome_size = cmd_bases(cmd, optarg); break;
case 'S': cmd_check(!conf_table_path,cmd); conf_table_path = optarg; break;
case 'M': cmd_check(use_missing_info_check,cmd); use_missing_info_check = false; break;
case 'P': cmd_check(!seed_with_unused_paths,cmd); seed_with_unused_paths = true; break;
case 'C':
cmd_check(min_cumul_confid < 0,cmd);
min_cumul_confid = cmd_udouble(cmd,optarg);
if(min_cumul_confid > 1) die("%s must be 0 <= x <= 1", cmd);
break;
case 'T':
cmd_check(min_step_confid < 0,cmd);
min_step_confid = cmd_udouble(cmd,optarg);
if(min_step_confid > 1) die("%s must be 0 <= x <= 1", cmd);
break;
case ':': /* BADARG */
case '?': /* BADCH getopt_long has already printed error */
die("`"CMD" contigs -h` for help. Bad option: %s", argv[optind-1]);
default: abort();
}
}
if(cmd_no_reseed && cmd_reseed)
cmd_print_usage("Cannot specify both -r and -R");
if(contig_limit && seed_with_unused_paths)
cmd_print_usage("Cannot combine --ncontigs with --use-seed-paths");
bool sample_with_replacement = cmd_reseed;
// Defaults
if(nthreads == 0) nthreads = DEFAULT_NTHREADS;
if(!seed_buf.len && !contig_limit && sample_with_replacement) {
cmd_print_usage("Please specify one or more of: "
"--no-reseed | --ncontigs | --seed <in.fa>");
}
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 ncols, ctx_max_kmers = 0, ctx_sum_kmers = 0;
graph_files_open(graph_paths, gfiles, num_gfiles,
&ctx_max_kmers, &ctx_sum_kmers);
// char *ctx_path = argv[optind];
//
// Open Graph file
//
// GraphFileReader gfile;
// memset(&gfile, 0, sizeof(GraphFileReader));
// graph_file_open(&gfile, ctx_path);
// Update colours in graph file - sample in 0, all others in 1
// never need more than two colours
ncols = gpath_load_sample_pop(gfiles, num_gfiles,
gpfiles.b, gpfiles.len, colour);
// Check for compatibility between graph files and path files
// pop_colour is colour 1
graphs_gpaths_compatible(gfiles, num_gfiles, gpfiles.b, gpfiles.len, 1);
if(!genome_size)
{
char nk_str[50];
if(ctx_max_kmers <= 0) die("Please pass --genome <G> if streaming");
genome_size = ctx_max_kmers;
ulong_to_str(genome_size, nk_str);
status("Taking number of kmers as genome size: %s", nk_str);
}
//
// Decide on memory
//
size_t bits_per_kmer, kmers_in_hash, graph_mem, path_mem, total_mem;
// 1 bit needed per kmer if we need to keep track of kmer usage
bits_per_kmer = sizeof(BinaryKmer)*8 + sizeof(Edges)*8 + sizeof(GPath*)*8 +
ncols + !sample_with_replacement;
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);
// Paths memory
size_t rem_mem = memargs.mem_to_use - MIN2(memargs.mem_to_use, graph_mem);
path_mem = gpath_reader_mem_req(gpfiles.b, gpfiles.len, ncols, rem_mem, 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");
// Total memory
total_mem = graph_mem + path_mem;
cmd_check_mem_limit(memargs.mem_to_use, total_mem);
// Load contig hist distribution from ctp files
ZeroSizeBuffer contig_hist;
memset(&contig_hist, 0, sizeof(contig_hist));
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),
&contig_hist);
}
// Calculate confidences, only for one colour
ContigConfidenceTable conf_table;
conf_table_alloc(&conf_table, 1);
conf_table_update_hist(&conf_table, 0, genome_size,
contig_hist.b, contig_hist.len);
if(conf_table_path != NULL) {
conf_table_save(&conf_table, conf_table_path);
}
zsize_buf_dealloc(&contig_hist);
//
// Output file if printing
//
FILE *fout = out_path ? futil_fopen_create(out_path, "w") : NULL;
// Allocate
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);
uint8_t *visited = NULL;
if(!sample_with_replacement)
visited = ctx_calloc(roundup_bits2bytes(db_graph.ht.capacity), 1);
// Load graph
LoadingStats stats = LOAD_STATS_INIT_MACRO;
GraphLoadingPrefs gprefs = {.db_graph = &db_graph,
.boolean_covgs = false,
.must_exist_in_graph = false,
.empty_colours = true};
for(i = 0; i < num_gfiles; i++) {
graph_load(&gfiles[i], gprefs, &stats);
graph_file_close(&gfiles[i]);
gprefs.empty_colours = false;
}
ctx_free(gfiles);
hash_table_print_stats(&db_graph.ht);
// Load path files
for(i = 0; i < gpfiles.len; i++) {
gpath_reader_load(&gpfiles.b[i], GPATH_DIE_MISSING_KMERS, &db_graph);
gpath_reader_close(&gpfiles.b[i]);
}
gpfile_buf_dealloc(&gpfiles);
AssembleContigStats assem_stats;
assemble_contigs_stats_init(&assem_stats);
assemble_contigs(nthreads, seed_buf.b, seed_buf.len,
contig_limit, visited,
use_missing_info_check, seed_with_unused_paths,
min_step_confid, min_cumul_confid,
fout, out_path, &assem_stats, &conf_table,
&db_graph, 0); // Sample always loaded into colour zero
if(fout && fout != stdout) fclose(fout);
assemble_contigs_stats_print(&assem_stats);
assemble_contigs_stats_destroy(&assem_stats);
conf_table_dealloc(&conf_table);
for(i = 0; i < seed_buf.len; i++)
seq_close(seed_buf.b[i]);
seq_file_ptr_buf_dealloc(&seed_buf);
ctx_free(visited);
db_graph_dealloc(&db_graph);
return EXIT_SUCCESS;
}
