void Scale::UpdateMatrix() const
{
m_Matrix.resize(m_Scale.size(), m_Scale.size());
for (Matrix::size_type row = 0; row < m_Matrix.size1(); row++)
{
for (Matrix::size_type column = 0; column < m_Matrix.size2(); column++)
{
if (row == column)
m_Matrix(row, column) = m_Scale(row);
else
m_Matrix(row, column) = 0.0f;
}
}
}
void
CDistances::ComputeMatrix(vector<CSequence>& query_data,
CHitList& hitlist,
SNCBIFullScoreMatrix& score_matrix,
Blast_KarlinBlk& karlin_blk)
{
int num_queries = query_data.size();
vector<double> self_score(num_queries);
// Get the self score of each sequence
x_GetSelfScores(query_data, hitlist, score_matrix,
self_score, karlin_blk);
// All sequences start out equally far from each other
m_Matrix.Resize(num_queries, num_queries);
m_Matrix.Set(1.0);
// Every pairwise alignment pulls the two sequences involved
// in the alignment closer together. This formulation of
// pairwise distances amounts to calculating the average
// per-letter distance between the two sequences. See
//
// Clarke et al, "Inferring Genome Trees by Using a Filter
// to Eliminate Phylogenetically Discordant Sequences and
// a Distance Matrix Based on Mean Normalized BLASTP Scores",
// Jour. Bacteriology Apr. 2002 pp 2072-2080
for (int i = 0; i < hitlist.Size(); i++) {
CHit *hit = hitlist.GetHit(i);
int j = hit->m_SeqIndex1;
int k = hit->m_SeqIndex2;
double align_score = karlin_blk.Lambda *
hit->m_Score - karlin_blk.logK;
_ASSERT(j < k);
m_Matrix(j,k) -= 0.5 * align_score *
(1.0 / self_score[j] + 1.0 / self_score[k]);
}
// Force distance matrix to be symmetric
for (int i = 0; i < num_queries; i++) {
m_Matrix(i, i) = 0.0;
for (int j = 0; j < i; j++) {
// clamp to zero any distances that are negative
// or too small
if (fabs(m_Matrix(j, i)) < 1e-6)
m_Matrix(j, i) = 0;
if (m_Matrix(j, i) < 0)
m_Matrix(j, i) = 0;
m_Matrix(i, j) = m_Matrix(j, i);
}
}
}
float Matrix::operator()(UINT row, UINT column) const
{
return m_Matrix(row, column);
}
float& Matrix::operator()(UINT row, UINT column)
{
return m_Matrix(row, column);
}
