/** Return the consensus sequence of the specified gap. */ static ContigPath fillGap(const Graph& g, const AmbPathConstraint& apConstraint, vector<bool>& seen, ofstream& outFasta) { if (opt::verbose > 1) cerr << "\n* " << get(vertex_name, g, apConstraint.source) << ' ' << apConstraint.dist << "N " << get(vertex_name, g, apConstraint.dest) << '\n'; Constraints constraints; constraints.push_back(Constraint(apConstraint.dest, apConstraint.dist + opt::distanceError)); ContigPaths solutions; unsigned numVisited = 0; constrainedSearch(g, apConstraint.source, constraints, solutions, numVisited); bool tooComplex = numVisited >= opt::maxCost; for (ContigPaths::iterator solIt = solutions.begin(); solIt != solutions.end(); solIt++) solIt->insert(solIt->begin(), apConstraint.source); ContigPath consensus; bool tooManySolutions = solutions.size() > opt::numBranches; if (tooComplex) { stats.tooComplex++; if (opt::verbose > 1) cerr << solutions.size() << " paths (too complex)\n"; } else if (tooManySolutions) { stats.numTooManySolutions++; if (opt::verbose > 1) cerr << solutions.size() << " paths (too many)\n"; } else if (solutions.empty()) { stats.numNoSolutions++; if (opt::verbose > 1) cerr << "no paths\n"; } else if (solutions.size() == 1) { if (opt::verbose > 1) cerr << "1 path\n" << solutions.front() << '\n'; stats.numMerged++; } else { assert(solutions.size() > 1); if (opt::verbose > 2) copy(solutions.begin(), solutions.end(), ostream_iterator<ContigPath>(cerr, "\n")); else if (opt::verbose > 1) cerr << solutions.size() << " paths\n"; consensus = align(g, solutions, outFasta); if (!consensus.empty()) { stats.numMerged++; // Mark contigs that are used in a consensus. markSeen(seen, solutions, true); if (opt::verbose > 1) cerr << consensus << '\n'; } else stats.notMerged++; } return consensus; }
/** Return an ambiguous path that agrees with all the given paths. */ static ContigPath constructAmbiguousPath(const Graph &g, const ContigNode& origin, const ContigPaths& paths) { assert(!paths.empty()); // Find the size of the smallest path. const ContigPath& firstSol = paths.front(); size_t min_len = firstSol.size(); for (ContigPaths::const_iterator it = paths.begin() + 1; it != paths.end(); ++it) min_len = min(min_len, it->size()); // Find the longest prefix. ContigPath vppath; size_t longestPrefix; bool commonPrefix = true; for (longestPrefix = 0; longestPrefix < min_len; longestPrefix++) { const ContigNode& common_path_node = firstSol[longestPrefix]; for (ContigPaths::const_iterator solIter = paths.begin(); solIter != paths.end(); ++solIter) { const ContigNode& pathnode = (*solIter)[longestPrefix]; if (pathnode != common_path_node) { // Found the longest prefix. commonPrefix = false; break; } } if (!commonPrefix) break; vppath.push_back(common_path_node); } // Find the longest suffix. ContigPath vspath; size_t longestSuffix; bool commonSuffix = true; for (longestSuffix = 0; longestSuffix < min_len-longestPrefix; longestSuffix++) { const ContigNode& common_path_node = firstSol[firstSol.size()-longestSuffix-1]; for (ContigPaths::const_iterator solIter = paths.begin(); solIter != paths.end(); ++solIter) { const ContigNode& pathnode = (*solIter)[solIter->size()-longestSuffix-1]; if (pathnode != common_path_node) { // Found the longest suffix. commonSuffix = false; break; } } if (!commonSuffix) break; vspath.push_back(common_path_node); } ContigPath out; out.reserve(vppath.size() + 1 + vspath.size()); out.insert(out.end(), vppath.begin(), vppath.end()); if (longestSuffix > 0) { const ContigPath& longestPath( *max_element(paths.begin(), paths.end(), ComparePathLength(g, origin))); unsigned length = calculatePathLength(g, origin, longestPath, longestPrefix, longestSuffix); // Account for the overlap on the right. int dist = length + getDistance(g, longestSuffix == longestPath.size() ? origin : *(longestPath.rbegin() + longestSuffix), *(longestPath.rbegin() + longestSuffix - 1)); // Add k-1 because it is the convention. int numN = dist + opt::k - 1; assert(numN > 0); out.push_back(ContigNode(numN, 'N')); out.insert(out.end(), vspath.rbegin(), vspath.rend()); } return out; }
/* Resolve ambiguous region using pairwise alignment * (Needleman-Wunsch) ('solutions' contain exactly two paths, from a * source contig to a dest contig) */ static ContigPath alignPair(const Graph& g, const ContigPaths& solutions, ofstream& out) { assert(solutions.size() == 2); assert(solutions[0].size() > 1); assert(solutions[1].size() > 1); assert(solutions[0].front() == solutions[1].front()); assert(solutions[0].back() == solutions[1].back()); ContigPath fstSol(solutions[0].begin()+1, solutions[0].end()-1); ContigPath sndSol(solutions[1].begin()+1, solutions[1].end()-1); if (fstSol.empty() || sndSol.empty()) { // This entire sequence may be deleted. const ContigPath& sol(fstSol.empty() ? sndSol : fstSol); assert(!sol.empty()); Sequence consensus(mergePath(g, sol)); assert(consensus.size() > opt::k - 1); string::iterator first = consensus.begin() + opt::k - 1; transform(first, consensus.end(), first, ::tolower); unsigned match = opt::k - 1; float identity = (float)match / consensus.size(); if (opt::verbose > 2) cerr << consensus << '\n'; if (opt::verbose > 1) cerr << identity << (identity < opt::identity ? " (too low)\n" : "\n"); if (identity < opt::identity) return ContigPath(); unsigned coverage = calculatePathProperties(g, sol).coverage; ContigNode u = outputNewContig(g, solutions, 1, 1, consensus, coverage, out); ContigPath path; path.push_back(solutions.front().front()); path.push_back(u); path.push_back(solutions.front().back()); return path; } Sequence fstPathContig(mergePath(g, fstSol)); Sequence sndPathContig(mergePath(g, sndSol)); if (fstPathContig == sndPathContig) { // These two paths have identical sequence. if (fstSol.size() == sndSol.size()) { // A perfect match must be caused by palindrome. typedef ContigPath::const_iterator It; pair<It, It> it = mismatch( fstSol.begin(), fstSol.end(), sndSol.begin()); assert(it.first != fstSol.end()); assert(it.second != sndSol.end()); assert(*it.first == get(vertex_complement, g, *it.second)); assert(equal(it.first+1, It(fstSol.end()), it.second+1)); if (opt::verbose > 1) cerr << "Palindrome: " << get(vertex_contig_name, g, *it.first) << '\n'; return solutions[0]; } else { // The paths are different lengths. cerr << PROGRAM ": warning: " "Two paths have identical sequence, which may be " "caused by a transitive edge in the overlap graph.\n" << '\t' << fstSol << '\n' << '\t' << sndSol << '\n'; return solutions[fstSol.size() > sndSol.size() ? 0 : 1]; } } unsigned minLength = min( fstPathContig.length(), sndPathContig.length()); unsigned maxLength = max( fstPathContig.length(), sndPathContig.length()); float lengthRatio = (float)minLength / maxLength; if (lengthRatio < opt::identity) { if (opt::verbose > 1) cerr << minLength << '\t' << maxLength << '\t' << lengthRatio << "\t(different length)\n"; return ContigPath(); } NWAlignment align; unsigned match = alignGlobal(fstPathContig, sndPathContig, align); float identity = (float)match / align.size(); if (opt::verbose > 2) cerr << align; if (opt::verbose > 1) cerr << identity << (identity < opt::identity ? " (too low)\n" : "\n"); if (identity < opt::identity) return ContigPath(); unsigned coverage = calculatePathProperties(g, fstSol).coverage + calculatePathProperties(g, sndSol).coverage; ContigNode u = outputNewContig(g, solutions, 1, 1, align.consensus(), coverage, out); ContigPath path; path.push_back(solutions.front().front()); path.push_back(u); path.push_back(solutions.front().back()); return path; }