/// \brief Remap the leaf indices of tree \a T to match the alignment \a A: check the result
///
/// \param A The alignment.
/// \param T The tree.
/// \param internal_sequences Should the resulting alignment have sequences for internal nodes on the tree?
///
void link(alignment& A,SequenceTree& T,bool internal_sequences)
{
check_names_unique(A);
// Later, might we WANT sub-branches???
if (has_sub_branches(T))
remove_sub_branches(T);
if (internal_sequences and not is_Cayley(T) and T.n_leaves() > 1) {
assert(has_polytomy(T));
throw myexception()<<"Cannot link a multifurcating tree to an alignment with internal sequences.";
}
//------ IF sequences < leaf nodes THEN complain ---------//
if (A.n_sequences() < T.n_leaves())
throw myexception()<<"Tree has "<<T.n_leaves()<<" leaves but Alignment only has "
<<A.n_sequences()<<" sequences.";
//----- IF sequences = leaf nodes THEN maybe add internal sequences.
else if (A.n_sequences() == T.n_leaves())
{
A = remap_A_indices(A,T);
if (internal_sequences)
{
add_internal_labels(T);
A = add_internal(A,T);
connect_leaf_characters(A,T);
}
}
//----- IF sequences > leaf nodes THEN maybe complain -------//
else if (A.n_sequences() > T.n_nodes())
throw myexception()<<"More alignment sequences ("<<A.n_sequences()<<") than tree nodes ("<<T.n_nodes()<<")!";
else if (A.n_sequences() < T.n_nodes())
throw myexception()<<"Fewer alignment sequences ("<<A.n_sequences()<<") than tree nodes ("<<T.n_nodes()<<")!";
else
{
A = remap_A_indices(A,T);
if (not internal_sequences)
A = chop_internal(A);
}
//---------- double-check that we have the right number of sequences ---------//
if (internal_sequences)
assert(A.n_sequences() == T.n_nodes());
else
assert(A.n_sequences() == T.n_leaves());
//----- Check that each alignment sequence maps to a corresponding name in the tree -----//
for(int i=0;i<A.n_sequences();i++)
assert(T.get_label(i) == A.seq(i).name);
//---- Check to see that internal nodes satisfy constraints ----//
check_alignment(A,T,internal_sequences);
}
vector<vector<int> > get_all_parsimony_letters(const alphabet& a, const vector<int>& letters, const SequenceTree& T,
const ublas::matrix<int>& cost)
{
int root = T.directed_branch(0).target();
ublas::matrix<int> n_muts(T.n_nodes(), a.size());
peel_n_mutations(a,letters,T,cost,n_muts, branches_toward_node(T,root) );
// get an order list of branches point away from the root;
vector<const_branchview> branches = branches_from_node(T,root);
std::reverse(branches.begin(),branches.end());
// Allocate space to store the letters for each node
vector<vector<int> > node_letters(T.n_nodes());
const unsigned A = a.size();
// choose the cheapest letters at the root
{
double m = row_min(n_muts,root);
for(int l=0;l<A;l++)
if (n_muts(root,l) <= m)
node_letters[root].push_back(l);
}
vector<double> temp(A);
for(int i=0;i<branches.size();i++)
{
int s = branches[i].source();
int t = branches[i].target();
vector<double> best(node_letters[s].size());
for(int j=0;j<node_letters[s].size();j++)
{
for(int l=0;l<A;l++)
temp[l] = n_muts(t,l)+cost(l,node_letters[s][j]);
best[j] = min(temp);
}
for(int l=0;l<A;l++)
{
bool is_best = false;
for(int j=0;j<node_letters[s].size() and not is_best;j++)
if (n_muts(t,l)+cost(l,node_letters[s][j]) <= best[j])
is_best=true;
if (is_best)
node_letters[t].push_back(l);
}
}
return node_letters;
}
/// \brief Remap the leaf indices of tree \a T to match the alignment \a A: check the result
///
/// \param A The alignment.
/// \param T The tree.
/// \param internal_sequences Should the resulting alignment have sequences for internal nodes on the tree?
///
void link(alignment& A,SequenceTree& T,bool internal_sequences)
{
check_names_unique(A);
// Later, might we WANT sub-branches???
if (has_sub_branches(T))
remove_sub_branches(T);
if (internal_sequences and not is_Cayley(T)) {
assert(has_polytomy(T));
throw myexception()<<"Cannot link a multifurcating tree to an alignment with internal sequences.";
}
//------ IF sequences < leaf nodes THEN complain ---------//
if (A.n_sequences() < T.n_leaves())
throw myexception()<<"Tree has "<<T.n_leaves()<<" leaves but Alignment only has "
<<A.n_sequences()<<" sequences.";
//----- IF sequences = leaf nodes THEN maybe add internal sequences.
else if (A.n_sequences() == T.n_leaves()) {
if (internal_sequences)
A = add_internal(A,T);
}
//----- IF sequences > leaf nodes THEN maybe complain -------//
else {
if (not internal_sequences) {
alignment A2 = chop_internal(A);
if (A2.n_sequences() == T.n_leaves()) {
A = A2;
}
else
throw myexception()<<"More alignment sequences than leaf nodes!";
}
else if (A.n_sequences() > T.n_nodes())
throw myexception()<<"More alignment sequences than tree nodes!";
else if (A.n_sequences() < T.n_nodes())
throw myexception()<<"Fewer alignment sequences than tree nodes!";
}
//---------- double-check that we have the right number of sequences ---------//
if (internal_sequences)
assert(A.n_sequences() == T.n_nodes());
else
assert(A.n_sequences() == T.n_leaves());
//----- Remap leaf indices for T onto A's leaf sequence indices -----//
remap_T_indices(T,A);
if (internal_sequences)
connect_leaf_characters(A,T);
//---- Check to see that internal nodes satisfy constraints ----//
check_alignment(A,T,internal_sequences);
}
alignment add_internal(alignment A,const SequenceTree& T)
{
// Complain if A and T don't correspond
if (A.n_sequences() != T.n_leaves())
throw myexception()<<"Number of sequence in alignment doesn't match number of leaves in tree"
<<"- can't add internal sequences";
// Add empty sequences
vector<sequence> S;
for(int i=T.n_leaves();i<T.n_nodes();i++)
{
sequence s;
if (T.label(i) == "")
throw myexception()<<"Adding internal sequences: Tree has missing internal node name!";
s.name = T.label(i);
S.push_back(s);
}
A.add_sequences(S);
return A;
}
void peel_n_mutations(const alphabet& a, const vector<int>& letters, const SequenceTree& T,
const ublas::matrix<B>& cost,ublas::matrix<B>& n_muts,
const vector<const_branchview>& branches)
{
const int A = a.size();
assert(letters.size() == T.n_leaves());
assert(cost.size1() == A);
assert(cost.size2() == A);
// we need a scratch row in the matrix
assert(n_muts.size1() == T.n_nodes());
assert(n_muts.size2() == A);
// compute the max cost -- is this approach a good idea?
// Well... this apparently doesn't work.
B max_cost = 0;
for(int i=0;i<A;i++)
for(int j=0;j<A;j++)
max_cost = std::max(cost(i,j)+1, max_cost);
// clear the length matrix.
for(int i=0;i<n_muts.size1();i++)
for(int j=0;j<n_muts.size2();j++)
n_muts(i,j)=0;
// set the leaf costs
for(int s=0;s<T.n_leaves();s++)
{
int L = letters[s];
if (a.is_letter_class(L))
for(int l=0;l<A;l++)
if (a.matches(l,L))
n_muts(s,l) = 0;
else
n_muts(s,l) = max_cost;
}
// compute the costs for letters at each node
for(int i=0;i<branches.size();i++)
{
int s = branches[i].source();
int t = branches[i].target();
// for each letter l of node target...
for(int l=0;l<A;l++)
{
// compute minimum treelength for data behind source.
B temp = n_muts(s,0)+cost(0,l);
for(int k=1;k<A;k++)
temp = min(temp, n_muts(s,k)+cost(k,l) );
// add it to treelengths for data behind target
n_muts(t,l) += temp;
}
}
}
accum_branch_lengths_same_topology(const SequenceTree& T)
:
n_samples(0),
n_matches(0),
Q(T),
m1(0.0, Q.n_branches()),
m2(0.0, Q.n_branches()),
n1(0.0, Q.n_nodes())
{}
void add_internal_labels(SequenceTree& T)
{
for(int i=0;i<T.n_nodes();i++)
if (T.node(i).is_internal_node())
{
if (T.get_label(i) == "")
T.set_label(i, string("A") + convertToString(i));
}
}
B n_mutations(const alphabet& a, const vector<int>& letters, const SequenceTree& T,const ublas::matrix<B>& cost)
{
int root = T.directed_branch(0).target();
vector<const_branchview> branches = branches_toward_node(T,root);
ublas::matrix<B> n_muts(T.n_nodes(), a.size());
return n_mutations(a,letters,T,cost,n_muts,branches);
}
vector<int> get_parsimony_letters(const alphabet& a, const vector<int>& letters, const SequenceTree& T,
const ublas::matrix<int>& cost)
{
int root = T.directed_branch(0).target();
ublas::matrix<int> n_muts(T.n_nodes(),a.size());
peel_n_mutations(a,letters,T,cost,n_muts, branches_toward_node(T,root) );
// get an order list of branches point away from the root;
vector<const_branchview> branches = branches_from_node(T,root);
std::reverse(branches.begin(),branches.end());
// Allocate space to store the letter for each node
vector<int> node_letters(T.n_nodes(),-1);
// choose the cheapest letter at the root
node_letters[root] = row_min(n_muts,root);
const unsigned A = a.size();
vector<double> temp(A);
for(int i=0;i<branches.size();i++)
{
int s = branches[i].source();
int t = branches[i].target();
int k = node_letters[s];
assert(k != -1);
for(int l=0;l<A;l++)
temp[l] = n_muts(t,l)+cost(l,k);
node_letters[t] = argmin(temp);
}
return node_letters;
}
/// \brief Re-index the leaves of tree \a T so that the labels have the same ordering as in \a A.
///
/// \param T The leaf-labelled tree.
/// \param A A multiple sequence alignment.
///
alignment remap_A_indices(alignment& A, const SequenceTree& T)
{
vector<string> labels = T.get_labels();
if (A.n_sequences() == T.n_leaves())
{
labels.resize(T.n_leaves());
}
else if (A.n_sequences() != T.n_nodes())
throw myexception()<<"Cannot map alignment onto tree:\n Alignment has "<<A.n_sequences()<<" sequences.\n Tree has "<<T.n_leaves()<<" leaves and "<<T.n_nodes()<<" nodes.";
for(int i=0;i<labels.size();i++)
if (labels[i] == "")
{
if (i<T.n_leaves())
throw myexception()<<"Tree has empty label for a leaf node: not allowed!";
else
throw myexception()<<"Alignment has internal node information, but tree has empty label for an internal node: not allowed!";
}
assert(A.n_sequences() == labels.size());
//----- Remap leaf indices for T onto A's leaf sequence indices -----//
try {
vector<int> mapping = compute_mapping(labels, sequence_names(A));
return reorder_sequences(A,mapping);
}
catch(const bad_mapping<string>& b)
{
bad_mapping<string> b2 = b;
b2.clear();
if (b.from == 0)
b2<<"Couldn't find sequence \""<<b2.missing<<"\" in alignment.";
else
b2<<"Alignment sequence '"<<b2.missing<<"' not found in the tree.";
throw b2;
}
}
B n_mutations(const alignment& A, const SequenceTree& T,const ublas::matrix<B>& cost)
{
const alphabet& a = A.get_alphabet();
vector<int> letters(T.n_leaves());
int root = T.directed_branch(0).target();
vector<const_branchview> branches = branches_toward_node(T,root);
ublas::matrix<B> n_muts(T.n_nodes(), a.size());
double tree_length = 0;
for(int c=0;c<A.length();c++) {
for(int i=0;i<T.n_leaves();i++)
letters[i] = A(c,i);
double length = n_mutations<B>(a,letters,T,cost,n_muts,branches);
tree_length += length;
}
return tree_length;
}
// mark nodes in T according to what node of Q they map to
vector<int> get_nodes_map(const SequenceTree& Q,const SequenceTree& T,
const vector<int>& branches_map)
{
assert(branches_map.size() == Q.n_branches() * 2);
vector<int> nodes_map(T.n_nodes(),-1);
// map nodes from T -> Q that are in both trees
for(int b=0;b<Q.n_branches();b++)
{
int Q_source = Q.branch(b).source();
int Q_target = Q.branch(b).target();
int b2 = branches_map[b];
int T_source = T.directed_branch(b2).source();
int T_target = T.directed_branch(b2).target();
if (nodes_map[T_source] == -1)
nodes_map[T_source] = Q_source;
else
assert(nodes_map[T_source] == Q_source);
if (nodes_map[T_target] == -1)
nodes_map[T_target] = Q_target;
else
assert(nodes_map[T_target] == Q_target);
}
// map the rest of the nodes from T -> Q
for(int i=Q.n_leaves();i<Q.n_nodes();i++)
{
unsigned D = Q[i].degree();
if (D <= 3) continue;
// get a branch of Q pointing into the node
const_branchview outside = *(Q[i].branches_in());
// get a branch of T pointing into the node
outside = T.directed_branch(branches_map[outside.name()]);
list<const_branchview> branches;
typedef list<const_branchview>::iterator list_iterator;
append(outside.branches_after(),branches);
for(list_iterator b = branches.begin() ; b != branches.end();)
{
int node = (*b).target();
if (nodes_map[node] == -1)
nodes_map[node] = i;
if (nodes_map[node] == i) {
append((*b).branches_after(),branches);
b++;
}
else {
list_iterator prev = b;
b++;
branches.erase(prev);
}
}
assert(branches.size() == D-3);
}
for(int i=0;i<nodes_map.size();i++)
assert(nodes_map[i] != -1);
return nodes_map;
}
int main(int argc,char* argv[])
{
try {
//----------- Parse command line ----------//
variables_map args = parse_cmd_line(argc,argv);
int skip = args["skip"].as<int>();
int max = -1;
if (args.count("max"))
max = args["max"].as<int>();
int subsample = args["sub-sample"].as<int>();
vector<string> prune;
if (args.count("prune")) {
string p = args["prune"].as<string>();
prune = split(p,',');
}
//----------- Read the topology -----------//
SequenceTree Q = load_T(args);
standardize(Q);
const int B = Q.n_branches();
const int N = Q.n_nodes();
vector<double> bf(B);
for(int b=0;b<bf.size();b++)
bf[b] = Q.branch(b).length();
//-------- Read in the tree samples --------//
if ( args.count("simple") ) {
accum_branch_lengths_ignore_topology A(Q);
scan_trees(std::cin,skip,subsample,max,prune,A);
for(int b=0;b<B;b++)
Q.branch(b).set_length(A.m1[b]);
cout<<Q.write_with_bootstrap_fraction(bf,true)<<endl;
exit(0);
}
accum_branch_lengths_same_topology A(Q);
try {
scan_trees(std::cin,skip,subsample,max,prune,A);
}
catch (std::exception& e)
{
if (args.count("safe"))
cout<<Q.write(false)<<endl;
std::cerr<<"tree-mean-lengths: Error! "<<e.what()<<endl;
exit(0);
}
if (log_verbose) std::cerr<<A.n_matches<<" out of "<<A.n_samples<<" trees matched the topology";
if (log_verbose) std::cerr<<" ("<<double(A.n_matches)/A.n_samples*100<<"%)"<<std::endl;
//------- Merge lengths and topology -------//
if (args.count("var")) {
for(int b=0;b<B;b++)
Q.branch(b).set_length(A.m2[b]);
cout<<Q;
exit(0);
}
else {
for(int b=0;b<B;b++)
Q.branch(b).set_length(A.m1[b]);
if (not args.count("no-node-lengths") and
not args.count("show-node-lengths")) {
for(int n=0;n<N;n++) {
int degree = Q[n].neighbors().size();
for(out_edges_iterator b = Q[n].branches_out();b;b++)
(*b).set_length((*b).length() + A.n1[n]/degree);
}
}
//------- Print Tree and branch lengths -------//
cout<<Q.write_with_bootstrap_fraction(bf,true)<<endl;
//------------ Print node lengths -------------//
if (args.count("show-node-lengths"))
for(int n=0;n<Q.n_nodes();n++) {
if (A.n1[n] > 0) {
cout<<"node "<<A.n1[n]<<endl;
int b = (*Q[n].branches_in()).name();
cout<<partition_from_branch(Q,b)<<endl;
}
}
}
}
catch (std::exception& e) {
std::cerr<<"tree-mean-lengths: Error! "<<e.what()<<endl;
exit(1);
}
return 0;
}