Alignment Sampler::mutate(const Alignment& aln,
double base_error,
double indel_error) {
if (base_error == 0 && indel_error == 0) return aln;
string bases = "ATGC";
uniform_real_distribution<double> rprob(0, 1);
uniform_int_distribution<int> rbase(0, 3);
Alignment mutaln;
for (size_t i = 0; i < aln.path().mapping_size(); ++i) {
auto& orig_mapping = aln.path().mapping(i);
Mapping* new_mapping = mutaln.mutable_path()->add_mapping();
*new_mapping->mutable_position() = orig_mapping.position();
// for each edit in the mapping
for (size_t j = 0; j < orig_mapping.edit_size(); ++j) {
auto& orig_edit = orig_mapping.edit(j);
auto new_edits = mutate_edit(orig_edit, make_pos_t(orig_mapping.position()),
base_error, indel_error,
bases, rprob, rbase);
for (auto& edit : new_edits) {
*new_mapping->add_edit() = edit;
}
}
}
// re-derive the alignment's sequence
mutaln = simplify(mutaln);
mutaln.set_sequence(alignment_seq(mutaln));
mutaln.set_name(aln.name());
return mutaln;
}
map<pos_t, char> Sampler::next_pos_chars(pos_t pos) {
map<pos_t, char> nexts;
// See if the node is cached (did we just visit it?)
pair<Node, bool> cached = node_cache.retrieve(id(pos));
if(!cached.second) {
// If it's not in the cache, put it in
cached.first = xgidx->node(id(pos));
node_cache.put(id(pos), cached.first);
}
Node& node = cached.first;
// if we are still in the node, return the next position and character
if (offset(pos) < node.sequence().size()-1) {
++get_offset(pos);
nexts[pos] = pos_char(pos);
} else {
// look at the next positions we could reach
if (!is_rev(pos)) {
// we are on the forward strand, the next things from this node come off the end
for (auto& edge : xgidx->edges_on_end(id(pos))) {
if (edge.from() == id(pos)) {
pos_t p = make_pos_t(edge.to(), edge.to_end(), 0);
nexts[p] = pos_char(p);
} else if (edge.from_start() && edge.to_end() && edge.to() == id(pos)) {
// doubly inverted, should be normalized to forward but we handle here for safety
pos_t p = make_pos_t(edge.from(), false, 0);
nexts[p] = pos_char(p);
}
}
} else {
// we are on the reverse strand, the next things from this node come off the start
for (auto& edge : xgidx->edges_on_start(id(pos))) {
if (edge.to() == id(pos)) {
pos_t p = make_pos_t(edge.from(), !edge.from_start(), 0);
nexts[p] = pos_char(p);
} else if (edge.from_start() && edge.to_end() && edge.from() == id(pos)) {
// doubly inverted, should be normalized to forward but we handle here for safety
pos_t p = make_pos_t(edge.to(), true, 0);
nexts[p] = pos_char(p);
}
}
}
}
return nexts;
}
pos_t Sampler::position(void) {
uniform_int_distribution<size_t> xdist(1, xgidx->seq_length);
size_t offset = xdist(rng);
id_t id = xgidx->node_at_seq_pos(offset);
uniform_int_distribution<size_t> flip(0, 1);
bool rev = forward_only ? false : flip(rng);
// 1-0 base conversion
size_t node_offset = offset - xgidx->node_start(id) - 1;
return make_pos_t(id, rev, node_offset);
}
vector<edge_t> find_edges_to_prune(const HandleGraph& graph, size_t k, size_t edge_max) {
// for each position on the forward and reverse of the graph
//unordered_set<edge_t> edges_to_prune;
vector<vector<edge_t> > edges_to_prune;
edges_to_prune.resize(get_thread_count());
graph.for_each_handle([&](const handle_t& h) {
// for the forward and reverse of this handle
// walk k bases from the end, so that any kmer starting on the node will be represented in the tree we build
for (auto handle_is_rev : { false, true }) {
//cerr << "###########################################" << endl;
handle_t handle = handle_is_rev ? graph.flip(h) : h;
list<walk_t> walks;
// for each position in the node, set up a kmer with that start position and the node end or kmer length as the end position
// determine next positions
id_t handle_id = graph.get_id(handle);
size_t handle_length = graph.get_length(handle);
string handle_seq = graph.get_sequence(handle);
for (size_t i = 0; i < handle_length; ++i) {
pos_t begin = make_pos_t(handle_id, handle_is_rev, handle_length);
pos_t end = make_pos_t(handle_id, handle_is_rev, min(handle_length, i+k));
walk_t walk = walk_t(offset(end)-offset(begin), begin, end, handle, 0);
if (walk.length < k) {
// are we branching over more than one edge?
size_t next_count = 0;
graph.follow_edges(walk.curr, false, [&](const handle_t& next) { ++next_count; });
graph.follow_edges(walk.curr, false, [&](const handle_t& next) {
if (next_count > 1 && edge_max == walk.forks) { // our next step takes us over the max
int tid = omp_get_thread_num();
edges_to_prune[tid].push_back(graph.edge_handle(walk.curr, next));
} else {
walks.push_back(walk);
auto& todo = walks.back();
todo.curr = next;
if (next_count > 1) {
++todo.forks;
}
}
});
} else {
walks.push_back(walk);
}
}
// now expand the kmers until they reach k
while (!walks.empty()) {
// first we check which ones have reached length k in the current handle; for each of these we run lambda and remove them from our list
auto walks_end = walks.end();
for (list<walk_t>::iterator q = walks.begin(); q != walks_end; ++q) {
auto& walk = *q;
// did we reach our target length?
if (walk.length >= k) {
q = walks.erase(q);
} else {
id_t curr_id = graph.get_id(walk.curr);
size_t curr_length = graph.get_length(walk.curr);
bool curr_is_rev = graph.get_is_reverse(walk.curr);
size_t take = min(curr_length, k-walk.length);
walk.end = make_pos_t(curr_id, curr_is_rev, take);
walk.length += take;
if (walk.length < k) {
// if not, we need to expand through the node then follow on
size_t next_count = 0;
graph.follow_edges(walk.curr, false, [&](const handle_t& next) { ++next_count; });
graph.follow_edges(walk.curr, false, [&](const handle_t& next) {
if (next_count > 1 && edge_max == walk.forks) { // our next step takes us over the max
int tid = omp_get_thread_num();
edges_to_prune[tid].push_back(graph.edge_handle(walk.curr, next));
} else {
walks.push_back(walk);
auto& todo = walks.back();
todo.curr = next;
if (next_count > 1) {
++todo.forks;
}
}
});
q = walks.erase(q);
} else {
// nothing, we'll remove it next time around
}
}
}
}
}
}, true);
uint64_t total_edges = 0;
for (auto& v : edges_to_prune) total_edges += v.size();
vector<edge_t> merged; merged.reserve(total_edges);
for (auto& v : edges_to_prune) {
merged.insert(merged.end(), v.begin(), v.end());
}
// duplicates are assumed to be dealt with externally
return merged;
}
std::vector<GaplessExtension> GaplessExtender::maximal_extensions(cluster_type& cluster, const std::string& sequence, bool cluster_is_sorted) const {
// Process the seeds in sorted order.
seed_type prev(0, make_pos_t(0, false, 0));
size_t prev_limit = 0; // Limit in the initial node.
if (!cluster_is_sorted) {
std::sort(cluster.begin(), cluster.end());
}
std::set<UnambiguousMatch> matches;
for (size_t i = 0; i < cluster.size(); i++) {
// Skip redundant seeds.
seed_type normalized = cluster[i];
size_t adjustment = std::min(static_cast<size_t>(offset(normalized.second)), normalized.first);
normalized.first -= adjustment;
get_offset(normalized.second) -= adjustment;
if (normalized == prev && offset(cluster[i].second) < prev_limit) {
continue;
}
prev = normalized;
// prev_limit is updated later when we match the first node.
// Match the initial node.
handle_t handle = GBWTGraph::node_to_handle(pos_to_gbwt(cluster[i].second));
UnambiguousMatch match {
cluster[i].first, cluster[i].first,
static_cast<size_t>(offset(cluster[i].second)), static_cast<size_t>(offset(cluster[i].second)),
this->graph->get_bd_state(handle),
{ handle }
};
std::pair<const char*, size_t> node_view = this->graph->get_sequence_view(handle);
match_forward(sequence, node_view, match);
prev_limit = match.node_limit;
match_backward(sequence, node_view, match);
// Match forward.
while (match.node_limit >= node_view.second && match.seq_limit < sequence.length()) {
bool extension = false, ambiguous = false;
gbwt::BidirectionalState successor;
this->graph->follow_paths(match.state, false, [&](const gbwt::BidirectionalState& next_state) -> bool {
if (ambiguous) {
return false;
}
if (next_state.empty()) {
return true;
}
handle_t next_handle = GBWTGraph::node_to_handle(next_state.forward.node);
if (this->graph->starts_with(next_handle, sequence[match.seq_limit])) {
if (extension) {
ambiguous = true;
return false;
} else {
extension = true;
successor = next_state;
handle = next_handle;
return true;
}
}
return true;
});
if (extension && !ambiguous) {
node_view = this->graph->get_sequence_view(handle);
match.seq_limit++;
match.node_limit = 1;
match.state = successor;
match.path.push_back(handle);
match_forward(sequence, node_view, match);
} else {
break;
}
}
// Match backward.
while (match.node_start == 0 && match.seq_start > 0) {
bool extension = false, ambiguous = false;
gbwt::BidirectionalState successor;
this->graph->follow_paths(match.state, true, [&](const gbwt::BidirectionalState& next_state) -> bool {
if (ambiguous) {
return false;
}
if (next_state.empty()) {
return true;
}
handle_t next_handle = GBWTGraph::node_to_handle(gbwt::Node::reverse(next_state.backward.node));
if (this->graph->ends_with(next_handle, sequence[match.seq_start - 1])) {
if (extension) {
ambiguous = true;
return false;
} else {
extension = true;
successor = next_state;
handle = next_handle;
return true;
}
}
return true;
});
if (extension && !ambiguous) {
node_view = this->graph->get_sequence_view(handle);
match.seq_start--;
match.node_start = node_view.second - 1;
match.state = successor;
match.path.insert(match.path.begin(), handle);
match_backward(sequence, node_view, match);
} else {
break;
}
}
if (!match.empty()) {
matches.insert(match);
}
}
// Convert the matches to GaplessExtension objects.
std::vector<GaplessExtension> result;
result.reserve(matches.size());
for(const UnambiguousMatch& match : matches) {
result.emplace_back(unambiguous_match_to_extension(match, *(this->graph), sequence));
}
return result;
}
GaplessExtension GaplessExtender::extend_seeds(cluster_type& cluster, const std::string& sequence, size_t max_mismatches, bool cluster_is_sorted) const {
GaplessMatch best_match {
max_mismatches + 1,
static_cast<size_t>(0), static_cast<size_t>(0),
static_cast<size_t>(0),
{ },
{ }
};
if (this->graph == nullptr) {
return match_to_extension(best_match, *(this->graph), sequence);
}
// Process the seeds in sorted order.
seed_type prev(0, make_pos_t(0, false, 0));
if (!cluster_is_sorted) {
std::sort(cluster.begin(), cluster.end());
}
for (size_t i = 0; i < cluster.size(); i++) {
// Start matching as early in the initial node as possible.
seed_type hit = cluster[i];
size_t adjustment = std::min(static_cast<size_t>(offset(hit.second)), hit.first);
get_offset(hit.second) -= adjustment;
hit.first -= adjustment;
if (hit == prev) {
continue; // This seed was redundant.
}
prev = hit;
// Match the initial node.
std::priority_queue<GaplessMatch> forward, backward;
{
handle_t handle = GBWTGraph::node_to_handle(pos_to_gbwt(hit.second));
GaplessMatch match {
static_cast<size_t>(0),
hit.first, hit.first,
static_cast<size_t>(offset(hit.second)),
this->graph->get_bd_state(handle),
{ },
{ }
};
match_forward(sequence, this->graph->get_sequence_view(handle), match.offset, match, best_match.score);
if (match.score >= best_match.score) {
continue;
} else {
match.path.push_back(handle);
}
if (match.limit >= sequence.length()) {
if (match.start == 0) {
best_match = match;
} else {
backward.push(match);
}
} else {
forward.push(match);
}
if (best_match.score == 0) {
return match_to_extension(best_match, *(this->graph), sequence);
}
}
// Match forward over all paths.
while (!forward.empty()) {
GaplessMatch curr = forward.top();
forward.pop();
this->graph->follow_paths(curr.state, false, [&](const gbwt::BidirectionalState& next_state) -> bool {
if (next_state.empty()) {
return true;
}
handle_t handle = GBWTGraph::node_to_handle(next_state.forward.node);
GaplessMatch next {
curr.score,
curr.start, curr.limit,
curr.offset,
next_state,
{ },
{ }
};
match_forward(sequence, this->graph->get_sequence_view(handle), 0, next, best_match.score);
if (next.score >= best_match.score) {
return true;
} else {
next.path.reserve(curr.path.size() + 1);
next.path.insert(next.path.end(), curr.path.begin(), curr.path.end());
next.path.push_back(handle);
next.mismatches.insert(next.mismatches.end(), curr.mismatches.begin(), curr.mismatches.end());
}
if (next.limit >= sequence.length()) {
if (next.start == 0) {
best_match = next;
} else {
backward.push(next);
}
} else {
forward.push(next);
}
return true;
});
if (best_match.score == 0) {
return match_to_extension(best_match, *(this->graph), sequence);
}
}
// Match backward over all paths.
while (!backward.empty()) {
GaplessMatch curr = backward.top();
backward.pop();
this->graph->follow_paths(curr.state, true, [&](const gbwt::BidirectionalState& next_state) -> bool {
if (next_state.empty()) {
return true;
}
handle_t handle = GBWTGraph::node_to_handle(gbwt::Node::reverse(next_state.backward.node));
GaplessMatch next {
curr.score,
curr.start, curr.limit,
curr.offset, // This will be replaced in match_backward().
next_state,
{ },
{ }
};
match_backward(sequence, this->graph->get_sequence_view(handle), next, best_match.score);
if (next.score >= best_match.score) {
return true;
} else {
next.path.reserve(curr.path.size() + 1);
next.path.push_back(handle);
next.path.insert(next.path.end(), curr.path.begin(), curr.path.end());
next.mismatches.insert(next.mismatches.end(), curr.mismatches.begin(), curr.mismatches.end());
}
if (next.start == 0) {
best_match = next;
} else {
backward.push(next);
}
return true;
});
if (best_match.score == 0) {
return match_to_extension(best_match, *(this->graph), sequence);
}
}
}
return match_to_extension(best_match, *(this->graph), sequence);
}