PathIndex::PathIndex(const xg::XG& index, const string& path_name, bool extract_sequence) {
// Make sure the path is present
assert(index.path_rank(path_name) != 0);
if (extract_sequence) {
// Constructor dispatch hack
*this = PathIndex(index.path(path_name), index);
} else {
*this = PathIndex(index.path(path_name));
}
}
void rectangle::simple_extend(thread_t& extension, xg::XG& graph, int delta_start = 0, int delta_end = 0) {
if(extension.size() > 0) {
xg::XG::ThreadMapping next_node = extension.back();
int64_t next_side = graph.id_to_rank(next_node.node_id) * 2 + next_node.is_reverse;
state.current_side = next_side;
}
state.range_start -= delta_start;
state.range_end -= delta_end;
}
bool check_for_edges(int64_t old_node_id, bool old_node_is_reverse, int64_t new_node_id,
bool new_node_is_reverse, xg::XG& index) {
// What edge are we following
Edge edge_taken = make_edge(old_node_id, old_node_is_reverse, new_node_id, new_node_is_reverse);
// Make sure we find it
bool edge_found = false;
vector<Edge> edges = new_node_is_reverse ? index.edges_on_end(new_node_id) : index.edges_on_start(new_node_id);
for(auto& edge : edges) {
// Look at every edge in order.
if(edges_equivalent(edge, edge_taken)) {
// If we found the edge we're taking, break.
edge_found = true;
break;
}
}
if(edge_found == false) { cerr << "did not find edge between" << old_node_id << " and " << new_node_id << endl;}
return edge_found;
}
void haplo_d::seeded_log_calculate_Is(xg::XG& graph) {
// Things which were calculated in the constructor:
// -- A's
// -- J for the top continuing and any new rectangle
// -- I for any new rectangle
// ostream& stream = cout;
vector<Edge> edges_out;
vector<Edge> edges_in;
for(int b = 1; b < cs.size(); b++) {
vector<rectangle>& prevAs = cs[b-1].S;
vector<rectangle>& currAs = cs[b].S;
// xg::XG::ThreadMapping lastnode;
// if(cs[b-1].bridge.size() == 0) {
// lastnode = cs[b-1].get_node();
// } else {
// lastnode = cs[b-1].bridge.back();
// }
// edges_out = lastnode.is_reverse ? graph.edges_on_start(lastnode.node_id) : graph.edges_on_end(lastnode.node_id);
edges_out = cs[b-1].get_last_node().is_reverse ? graph.edges_on_start(cs[b-1].get_last_node().node_id) : graph.edges_on_end(cs[b-1].get_last_node().node_id);
edges_in = cs[b].get_node().is_reverse ? graph.edges_on_end(cs[b].get_node().node_id) : graph.edges_on_start(cs[b].get_node().node_id);
bool new_threads = (prevAs[0].next == 1);
// make sure that there is at least one rectangle here
if(prevAs.size() == 0) {
cerr << "[vg haplo error] no consistent haplotypes at node " << cs[b-1].get_node().node_id << endl;
} else if(prevAs.size() == 1) {
currAs.back().I = currAs.back().J;
// currAs has size at most 2
if(currAs.size() == 2) {
currAs[0].I = currAs[0].J - currAs[1].J;
}
} else if(prevAs.size() >= 2) {
// We're going to have to extend, so let's grab the next node
XG::ThreadMapping next_node = cs[b].get_node();
// Let's also grab the nodes which we'll skip over between this and the last node
thread_t extension = cs[b-1].bridge;
extension.push_back(next_node);
// if J = 0 for a rectangle, then J must be 0 for all older rectangles
if(currAs.back().J == 0) {
currAs.pop_back();
} else {
int deltaJ = prevAs[0].J - currAs[prevAs[0].next].J;
if(deltaJ == 0) {
// cerr << b << ", deltaJ = 0" << endl;
currAs[prevAs[0].next].I = prevAs[0].I;
int delta_start = prevAs[0].state.range_start - currAs[prevAs[0].next].state.range_start;
int delta_end = prevAs[0].state.range_end - currAs[prevAs[0].next].state.range_end;
for(int a = 1; a < prevAs.size(); a++) {
rectangle new_rect = prevAs[a];
new_rect.simple_extend(extension, graph, delta_start, delta_end);
new_rect.prev = a;
currAs.push_back(new_rect);
prevAs[a].next = currAs.size()-1;
}
} else {
vector<int> previously_big;
int big_cutoff = 400;
for(int i = 1; i < prevAs.size(); i++) {
if(prevAs[i].I >= big_cutoff) {
previously_big.push_back(i);
}
}
// cerr << "made big list, it's size " << previously_big.size() << endl;
vector<rectangle> big_rectangles;
vector<int> big_deltas;
vector<int> big_Js;
for(int i = 0; i < previously_big.size(); i++) {
big_rectangles.push_back(prevAs[previously_big[i]]);
int Jbig = big_rectangles.back().get_next_J(extension,graph, edges_in, edges_out);
// cerr << Jbig << "\t" << flush;
big_Js.push_back(Jbig);
big_deltas.push_back(prevAs[previously_big[i]].J - Jbig);
if(Jbig == 0) {
break;
}
}
// cerr << endl;
// cerr << "collected attributes of big rectangles, " << big_Js.size() << " are nonempty" << endl;
if(big_Js.size() > 0) {
int Aabove = 0;
int Jabove = currAs[prevAs[0].next].J;
int dJabove = prevAs[0].J - currAs[prevAs[0].next].J;
for(int i = 0; i < previously_big.size(); i++) {
if(big_Js[i] == Jabove) {
// all rectangles between are actually empty
prevAs[currAs.back().prev].next = -1;
currAs.pop_back();
} else {
binaryI(graph, extension, b, Aabove, previously_big[i], dJabove, big_deltas[i], Jabove, big_Js[i], 0, edges_in, edges_out);
}
Aabove = previously_big[i];
Jabove = big_Js[i];
dJabove = big_deltas[i];
if(big_Js[i] == 0) {
// Don't build smaller rectangles
// Don't add this rectangle
break;
} else {
big_rectangles[i].prev = previously_big[i];
currAs.push_back(big_rectangles[i]);
prevAs[previously_big[i]].next = currAs.size()-1;
}
}
if(big_Js.back() != 0) {
binaryI(graph, extension, b, previously_big.back(), prevAs.size(), big_deltas.back(), 0, big_Js.back(), 0, 0, edges_in, edges_out);
}
} else {
binaryI(graph, extension, b, 0, prevAs.size(), deltaJ, 0, currAs[prevAs[0].next].J, 0, 0, edges_in, edges_out);
}
for(int a = 0; a < currAs.size() - 1; a++) {
currAs[a].I = currAs[a].J - currAs[a+1].J;
}
currAs.back().I = currAs.back().J;
}
}
}
}
}
void haplo_d::log_calculate_Is(xg::XG& graph) {
// Things which were calculated in the constructor:
// -- A's
// -- J for the top continuing and any new rectangle
// -- I for any new rectangle
// ostream& stream = cout;
vector<Edge> edges_out;
vector<Edge> edges_in;
for(int b = 1; b < cs.size(); b++) {
// xg::XG::ThreadMapping lastnode;
// if(cs[b-1].bridge.size() == 0) {
// lastnode = cs[b-1].get_node();
// } else {
// lastnode = cs[b-1].bridge.back();
// }
edges_out = cs[b-1].get_last_node().is_reverse ? graph.edges_on_start(cs[b-1].get_last_node().node_id) : graph.edges_on_end(cs[b-1].get_last_node().node_id);
// edges_out = lastnode.is_reverse ? graph.edges_on_start(lastnode.node_id) : graph.edges_on_end(lastnode.node_id);
edges_in = cs[b].get_node().is_reverse ? graph.edges_on_end(cs[b].get_node().node_id) : graph.edges_on_start(cs[b].get_node().node_id);
vector<rectangle>& prevAs = cs[b-1].S;
vector<rectangle>& currAs = cs[b].S;
bool new_threads = (prevAs[0].next == 1);
// make sure that there is at least one rectangle here
if(prevAs.size() == 0) {
cerr << "[vg haplo error] no consistent haplotypes at node " << cs[b-1].get_node().node_id << endl;
} else if(prevAs.size() == 1) {
currAs.back().I = currAs.back().J;
// currAs has size at most 2
if(currAs.size() == 2) {
currAs[0].I = currAs[0].J - currAs[1].J;
}
} else if(prevAs.size() >= 2) {
// We're going to have to extend, so let's grab the next node
XG::ThreadMapping next_node = cs[b].get_node();
// Let's also grab the nodes which we'll skip over between this and the last node
thread_t extension = cs[b-1].bridge;
extension.push_back(next_node);
// if J = 0 for a rectangle, then J must be 0 for all older rectangles
if(currAs.back().J == 0) {
currAs.pop_back();
} else {
int deltaJ = prevAs[0].J - currAs[prevAs[0].next].J;
if(deltaJ == 0) {
// cerr << b << ", deltaJ = 0" << endl;
currAs[prevAs[0].next].I = prevAs[0].I;
int delta_start = prevAs[0].state.range_start - currAs[prevAs[0].next].state.range_start;
int delta_end = prevAs[0].state.range_end - currAs[prevAs[0].next].state.range_end;
for(int a = 1; a < prevAs.size(); a++) {
rectangle new_rect = prevAs[a];
new_rect.simple_extend(extension, graph, delta_start, delta_end);
new_rect.prev = a;
currAs.push_back(new_rect);
prevAs[a].next = currAs.size()-1;
}
} else {
// binaryI(XG&, thread_t, b, atop, abott, dJtop, dJbott, Jtop, Jbott, indent level)
binaryI(graph, extension, b, 0, prevAs.size(), deltaJ, 0, currAs[prevAs[0].next].J, 0, 0, edges_in, edges_out);
for(int a = 0; a < currAs.size() - 1; a++) {
currAs[a].I = currAs[a].J - currAs[a+1].J;
}
currAs.back().I = currAs.back().J;
}
}
}
}
}
void rectangle::simple_extend(xg::XG::ThreadMapping next_node, xg::XG& graph, int delta_start = 0, int delta_end = 0) {
int64_t next_side = graph.id_to_rank(next_node.node_id) * 2 + next_node.is_reverse;
state.current_side = next_side;
state.range_start -= delta_start;
state.range_end -= delta_end;
}
haplo_d recombine_arms(haplo_d& left, haplo_d& right, int left_cut, int right_join, xg::XG& graph) {
haplo_d to_return;
if(!right.has_joining_node(right_join)) {
return to_return;
} else {
vector<rectangle*> boundary = right.trace_strip(right_join, 0, -1);
rectangle rect = left.cs[left_cut].S[0];
thread_t extension = left.cs[left_cut].bridge;
int lastJ = rect.J;
vector<int> boundaryDeltas;
vector<int> boundaryJs;
for(int i = 0; i < boundary.size(); i++) {
to_return.cs.push_back(right.cs[right_join + i].cs_shell());
extension.push_back(to_return.cs[i].get_node());
int new_J = rect.get_next_J(extension,graph);
boundaryDeltas.push_back(new_J - lastJ);
if(new_J > 0) {
boundaryJs.push_back(new_J);
} else {
break;
}
if(boundary[i]->J - new_J > 0) {
rectangle joiners;
if(i > 0) {
joiners.prev = 0;
to_return.cs[i-1].S[0].next = 0;
} else {
joiners.prev = -1;
}
joiners.J = boundary[i]->J;
joiners.I = boundary[i]->J - new_J;
to_return.cs[i].S.push_back(joiners);
}
if(new_J > 0) {
rectangle continuing;
continuing.J = new_J;
if(i > 0) {
continuing.prev = to_return.cs[i-1].S.size() - 1;
to_return.cs[i-1].S.back().next = to_return.cs[i].S.size();
}
to_return.cs[i].S.push_back(continuing);
}
lastJ = new_J;
extension = left.cs[left_cut].bridge;
}
for(int i = 0; i < left.cs[left_cut].S.size(); i++) {
//build first column
}
vector<Edge> edges_out;
vector<Edge> edges_in;
for(int i = 1; i < boundaryJs.size(); i++) {
edges_out = to_return.cs[i-1].get_last_node().is_reverse ? graph.edges_on_start(to_return.cs[i-1].get_last_node().node_id) : graph.edges_on_end(to_return.cs[i-1].get_last_node().node_id);
edges_in = to_return.cs[i].get_node().is_reverse ? graph.edges_on_end(to_return.cs[i].get_node().node_id) : graph.edges_on_start(to_return.cs[i].get_node().node_id);
to_return.binaryI(graph, extension, i, to_return.cs[i].S.back().prev, to_return.cs[i-1].S.size(), boundaryDeltas[i], 0, boundaryJs[i], 0, 0, edges_in, edges_out);
for(int a = 0; a < to_return.cs[i].S.size() - 1; a++) {
to_return.cs[i].S[a].I = to_return.cs[i].S[a].J - to_return.cs[i].S[a+1].J;
}
to_return.cs[i].S.back().I = to_return.cs[i].S.back().J;
}
}
}
//TODO: translate this
void haplo_d::initialize_skeleton(thread_t& t, pair<int,int> interval, cross_section& prevAs, xg::XG& graph) {
rectangle rect;
int new_height;
int last_height = prevAs.S[0].J;
bool add_rectangle;
bool add_A;
//TODO: fix this
int width = 0;
for(int i = interval.first; i <= interval.second; i++) {
// Count the number of base pairs since the last entry or exit node
width += graph.node_length(t[i-1].node_id);
new_height = graph.node_height(t[i]);
if(cs.back().S.size() != 0) {
if(i == interval.first) {
prevAs.S[0];
} else {
rect = cs.back().S[0];
}
rect.J = rect.get_next_J(t[i],graph); // step this strip forward
// Did any threads leave?
if(last_height > rect.J) {
add_A = 1;
}
// Are there any threads here which didn't come from the previous node?
if(rect.J < new_height) {
add_rectangle = 1;
add_A = 1;
}
// This is an entry or exit node, add a cross-section to the vector of
// cross-sections (which corresponds to the "A" set in the theory doc)
if(add_A) {
cs.back().width = width;
width = 0;
cs.push_back(cross_section(new_height,i,t[i]));
} else {
// This isn't a node where anything leaves or joins, let's skip over it
cs.back().bridge.push_back(t[i]);
for (size_t a = 0; a < cs.back().S.size(); a++) {
cs.back().S[a].extend(t[i],graph);
}
}
// This is an entry node; we also need a new rectangle corresponding to the
// new strip. We need to do this *before* we populate since cross_sections
// arrange rectangles newest -> oldest
// NB that add_rectangle implies add_A
if(add_rectangle) {
rectangle new_rect;
new_rect.extend(t[i],graph);
new_rect.J = new_height;
cs.back().height = new_rect.J;
cs.back().S.push_back(new_rect);
cs.back().S.back().I = new_rect.J - rect.J;
}
if(add_A) {
int b = cs.size()-1;
if(rect.J > 0) {
cs[b].S.push_back(rect);
cs[b].S.back().prev = 0;
cs[b-1].S[0].next = cs[b].S.size()-1;
}
}
last_height = new_height;
add_A = 0;
add_rectangle = 0;
} else {
cs.back().width = width;
width = 0;
cs.push_back(cross_section(new_height,i,t[i]));
if(new_height > 0) {
rectangle new_rect;
new_rect.extend(t[i],graph);
new_rect.J = new_height;
cs.back().height = new_rect.J;
cs.back().S.push_back(new_rect);
cs.back().S.back().I = new_rect.J - rect.J;
}
}
}
if(cs.size() == 1) {
cs.back().width = width;
}
cs.back().width += graph.node_length(t.back().node_id) - 1;
for(int i = 0; i < cs.size(); i++) {
tot_width += cs[i].width;
}
}
PathIndex::PathIndex(const Path& path, const xg::XG& index) {
// Trace the given path in the given XG graph, collecting sequence
// We're going to build the sequence string
std::stringstream seq_stream;
// What base are we at in the path?
size_t path_base = 0;
// What was the last rank? Ranks must always go up.
int64_t last_rank = -1;
for (size_t i = 0; i < path.mapping_size(); i++) {
auto& mapping = path.mapping(i);
if (!by_id.count(mapping.position().node_id())) {
// This is the first time we have visited this node in the path.
// Add in a mapping.
by_id[mapping.position().node_id()] =
std::make_pair(path_base, mapping.position().is_reverse());
#ifdef debug
#pragma omp critical (cerr)
std::cerr << "Node " << mapping.position().node_id() << " rank " << mapping.rank()
<< " starts at base " << path_base << " with "
<< index.node_sequence(mapping.position().node_id()) << std::endl;
#endif
// Make sure ranks are monotonically increasing along the path, or
// unset.
assert(mapping.rank() > last_rank || (mapping.rank() == 0 && last_rank == 0));
last_rank = mapping.rank();
}
// Say that this node appears here along the reference in this
// orientation.
by_start[path_base] = NodeSide(mapping.position().node_id(), mapping.position().is_reverse());
// Remember that occurrence by node ID.
node_occurrences[mapping.position().node_id()].push_back(by_start.find(path_base));
// Find the node's sequence
std::string node_sequence = index.node_sequence(mapping.position().node_id());
while(path_base == 0 && node_sequence.size() > 0 &&
(node_sequence[0] != 'A' && node_sequence[0] != 'T' && node_sequence[0] != 'C' &&
node_sequence[0] != 'G' && node_sequence[0] != 'N')) {
// If the path leads with invalid characters (like "X"), throw them
// out when computing path positions.
// TODO: this is a hack to deal with the debruijn-brca1-k63 graph,
// which leads with an X.
#pragma omp critical (cerr)
std::cerr << "Warning: dropping invalid leading character "
<< node_sequence[0] << " from node " << mapping.position().node_id()
<< std::endl;
node_sequence.erase(node_sequence.begin());
}
if (mapping.position().is_reverse()) {
// Put the reverse sequence in the path
seq_stream << reverse_complement(node_sequence);
} else {
// Put the forward sequence in the path
seq_stream << node_sequence;
}
// Whether we found the right place for this node in the reference or
// not, we still need to advance along the reference path. We assume the
// whole node (except any leading bogus characters) is included in the
// path (since it sort of has to be, syntactically, unless it's the
// first or last node).
path_base += node_sequence.size();
// TODO: handle leading bogus characters in calls on the first node.
}
// Record the length of the last mapping's node, since there's no next mapping to work it out from
last_node_length = path.mapping_size() > 0 ?
index.node_length(path.mapping(path.mapping_size() - 1).position().node_id()) :
0;
// Create the actual reference sequence we will use
sequence = seq_stream.str();
#ifdef debug
// Announce progress.
#pragma omp critical (cerr)
std::cerr << "Traced " << path_base << " bp path." << std::endl;
if (sequence.size() < 100) {
#pragma omp critical (cerr)
std::cerr << "Sequence: " << sequence << std::endl;
}
#endif
}