CNWAligner::TScore CPSSMAligner::x_AlignPSSM(SAlignInOut* data)
{
const size_t N1 = data->m_len1 + 1;
const size_t N2 = data->m_len2 + 1;
vector<TScore> stl_rowV (N2), stl_rowF(N2);
TScore* rowV = &stl_rowV[0];
TScore* rowF = &stl_rowF[0];
TScore* pV = rowV - 1;
const TScore** pssm_row = m_Pssm1 + data->m_offset1 - 1;
const char* seq2 = m_Seq2 + data->m_offset2 - 1;
m_terminate = false;
if(m_prg_callback) {
m_prg_info.m_iter_total = N1*N2;
m_prg_info.m_iter_done = 0;
if(m_terminate = m_prg_callback(&m_prg_info)) {
return 0;
}
}
TScore wg1L = m_Wg;
TScore wg1R = m_Wg;
TScore wg2L = m_Wg;
TScore wg2R = m_Wg;
TScore ws1L = m_Ws;
TScore ws1R = m_Ws;
TScore ws2L = m_Ws;
TScore ws2R = m_Ws;
if (data->m_offset1 == 0) {
if (data->m_esf_L1) {
wg1L = ws1L = 0;
}
else {
wg1L = m_StartWg;
ws1L = m_StartWs;
}
}
if (m_SeqLen1 == data->m_offset1 + data->m_len1) {
if (data->m_esf_R1) {
wg1R = ws1R = 0;
}
else {
wg1R = m_EndWg;
ws1R = m_EndWs;
}
}
if (data->m_offset2 == 0) {
if (data->m_esf_L2) {
wg2L = ws2L = 0;
}
else {
wg2L = m_StartWg;
ws2L = m_StartWs;
}
}
if (m_SeqLen2 == data->m_offset2 + data->m_len2) {
if (data->m_esf_R2) {
wg2R = ws2R = 0;
}
else {
wg2R = m_EndWg;
ws2R = m_EndWs;
}
}
TScore wgleft1 = wg1L;
TScore wsleft1 = ws1L;
TScore wg1 = m_Wg, ws1 = m_Ws;
// index calculation: [i,j] = i*n2 + j
CBacktraceMatrix4 backtrace_matrix (N1 * N2);
backtrace_matrix.SetAt(0, 0);
// first row
size_t k;
rowV[0] = wgleft1;
for (k = 1; k < N2; k++) {
rowV[k] = pV[k] + wsleft1;
rowF[k] = kInfMinus;
backtrace_matrix.SetAt(k, kMaskE | kMaskEc);
}
backtrace_matrix.Purge(k);
rowV[0] = 0;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
m_terminate = m_prg_callback(&m_prg_info);
}
// recurrences
TScore wgleft2 = wg2L;
TScore wsleft2 = ws2L;
TScore V = rowV[N2 - 1];
TScore V0 = wgleft2;
TScore E, G, n0;
unsigned char tracer;
size_t i, j;
for(i = 1; i < N1 && !m_terminate; ++i) {
V = V0 += wsleft2;
E = kInfMinus;
backtrace_matrix.SetAt(k++, kMaskFc);
if(i == N1 - 1) {
wg1 = wg1R;
ws1 = ws1R;
}
TScore wg2 = m_Wg, ws2 = m_Ws;
for (j = 1; j < N2; ++j, ++k) {
G = pV[j] + pssm_row[i][(unsigned char)seq2[j]];
pV[j] = V;
n0 = V + wg1;
if(E >= n0) {
E += ws1;
tracer = kMaskEc;
}
else {
E = n0 + ws1;
tracer = 0;
}
if(j == N2 - 1) {
wg2 = wg2R;
ws2 = ws2R;
}
n0 = rowV[j] + wg2;
if(rowF[j] >= n0) {
rowF[j] += ws2;
tracer |= kMaskFc;
}
else {
rowF[j] = n0 + ws2;
}
if (E >= rowF[j]) {
if(E >= G) {
V = E;
tracer |= kMaskE;
}
else {
V = G;
tracer |= kMaskD;
}
} else {
if(rowF[j] >= G) {
V = rowF[j];
}
else {
V = G;
tracer |= kMaskD;
}
}
backtrace_matrix.SetAt(k, tracer);
}
pV[j] = V;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
if(m_terminate = m_prg_callback(&m_prg_info)) {
break;
}
}
}
backtrace_matrix.Purge(k);
if(!m_terminate) {
x_DoBackTrace(backtrace_matrix, data);
}
return V;
}
CNWAligner::TScore CPSSMAligner::x_AlignProfile(SAlignInOut* data)
{
const size_t N1 = data->m_len1 + 1;
const size_t N2 = data->m_len2 + 1;
vector<double> stl_rowV (N2), stl_rowF(N2);
double* rowV = &stl_rowV[0];
double* rowF = &stl_rowF[0];
double* pV = rowV - 1;
const double** freq1_row = m_Freq1 + data->m_offset1 - 1;
const double** freq2_row = m_Freq2 + data->m_offset2 - 1;
m_terminate = false;
if(m_prg_callback) {
m_prg_info.m_iter_total = N1*N2;
m_prg_info.m_iter_done = 0;
if(m_terminate = m_prg_callback(&m_prg_info)) {
return 0;
}
}
TScore wg1L = m_Wg;
TScore wg1R = m_Wg;
TScore wg2L = m_Wg;
TScore wg2R = m_Wg;
TScore ws1L = m_Ws;
TScore ws1R = m_Ws;
TScore ws2L = m_Ws;
TScore ws2R = m_Ws;
if (data->m_offset1 == 0) {
if (data->m_esf_L1) {
wg1L = ws1L = 0;
}
else {
wg1L = m_StartWg;
ws1L = m_StartWs;
}
}
if (m_SeqLen1 == data->m_offset1 + data->m_len1) {
if (data->m_esf_R1) {
wg1R = ws1R = 0;
}
else {
wg1R = m_EndWg;
ws1R = m_EndWs;
}
}
if (data->m_offset2 == 0) {
if (data->m_esf_L2) {
wg2L = ws2L = 0;
}
else {
wg2L = m_StartWg;
ws2L = m_StartWs;
}
}
if (m_SeqLen2 == data->m_offset2 + data->m_len2) {
if (data->m_esf_R2) {
wg2R = ws2R = 0;
}
else {
wg2R = m_EndWg;
ws2R = m_EndWs;
}
}
TScore wgleft1 = wg1L;
TScore wsleft1 = ws1L;
TScore wg1 = m_Wg, ws1 = m_Ws;
// index calculation: [i,j] = i*n2 + j
CBacktraceMatrix4 backtrace_matrix (N1 * N2);
// first row
size_t k = 1;
if (N2 > 1) {
rowV[0] = wgleft1 * (1.0 - freq2_row[1][0]);
for (k = 1; k < N2; k++) {
rowV[k] = pV[k] + wsleft1;
rowF[k] = kInfMinus;
backtrace_matrix.SetAt(k, kMaskE | kMaskEc);
}
backtrace_matrix.Purge(k);
}
rowV[0] = 0;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
m_terminate = m_prg_callback(&m_prg_info);
}
// recurrences
TScore wgleft2 = wg2L;
TScore wsleft2 = ws2L;
double V = rowV[N2 - 1];
double V0 = 0;
double E, G, n0;
unsigned char tracer;
if (N1 > 1)
V0 = wgleft2 * (1.0 - freq1_row[1][0]);
size_t i, j;
for(i = 1; i < N1 && !m_terminate; ++i) {
V = V0 += wsleft2;
E = kInfMinus;
backtrace_matrix.SetAt(k++, kMaskFc);
if(i == N1 - 1) {
wg1 = wg1R;
ws1 = ws1R;
}
TScore wg2 = m_Wg, ws2 = m_Ws;
for (j = 1; j < N2; ++j, ++k) {
if(j == N2 - 1) {
wg2 = wg2R;
ws2 = ws2R;
}
const double *profile1 = freq1_row[i];
const double *profile2 = freq2_row[j];
const double scaled_wg1 = wg1 * (1.0 - profile2[0]);
const double scaled_ws1 = ws1;
const double scaled_wg2 = wg2 * (1.0 - profile1[0]);
const double scaled_ws2 = ws2;
double accum = 0.0, sum = 0.0;
int num_zeros1 = 0, num_zeros2 = 0;
double diff_freq1[kPSSM_ColumnSize];
double diff_freq2[kPSSM_ColumnSize];
// separate the residue frequencies into two components:
// a component that is the same for both columns, and
// a component that is different. The all-against-all
// score computation only takes place on the components
// that are different, so this will assign a higher score
// to more similar frequency columns
//
// Begin by separating out the common portion of each
// profile
for (int m = 1; m < kPSSM_ColumnSize; m++) {
if (profile1[m] < profile2[m]) {
accum += profile1[m] * m_DScoreMatrix[m][m];
diff_freq1[m] = 0.0;
diff_freq2[m] = profile2[m] - profile1[m];
num_zeros1++;
}
else {
accum += profile2[m] * m_DScoreMatrix[m][m];
diff_freq1[m] = profile1[m] - profile2[m];
diff_freq2[m] = 0.0;
num_zeros2++;
}
}
// normalize difference for profile with smaller gap
if (profile1[0] <= profile2[0]) {
for (int m = 1; m < kPSSM_ColumnSize; m++)
sum += diff_freq1[m];
} else {
for (int m = 1; m < kPSSM_ColumnSize; m++)
sum += diff_freq2[m];
}
if (sum > 0) {
sum = 1.0 / sum;
if (profile1[0] <= profile2[0]) {
for (int m = 1; m < kPSSM_ColumnSize; m++)
diff_freq1[m] *= sum;
} else {
for (int m = 1; m < kPSSM_ColumnSize; m++)
diff_freq2[m] *= sum;
}
// Add in the cross terms (not counting gaps).
// Note that the following assumes a symmetric
// score matrix
if (num_zeros1 > num_zeros2) {
for (int m = 1; m < kPSSM_ColumnSize; m++) {
if (diff_freq1[m] > 0) {
sum = 0.0;
double *matrix_row = m_DScoreMatrix[m];
for (int n = 1; n < kPSSM_ColumnSize; n++) {
sum += diff_freq2[n] * matrix_row[n];
}
accum += diff_freq1[m] * sum;
}
}
} else {
for (int m = 1; m < kPSSM_ColumnSize; m++) {
if (diff_freq2[m] > 0) {
sum = 0.0;
double *matrix_row = m_DScoreMatrix[m];
for (int n = 1; n < kPSSM_ColumnSize; n++) {
sum += diff_freq1[n] * matrix_row[n];
}
accum += diff_freq2[m] * sum;
}
}
}
}
G = pV[j] + accum * m_FreqScale +
profile1[0] * m_Ws * (1-profile2[0]) +
profile2[0] * m_Ws * (1-profile1[0]);
pV[j] = V;
n0 = V + scaled_wg1;
if(E >= n0) {
E += scaled_ws1; // continue the gap
tracer = kMaskEc;
}
else {
E = n0 + scaled_ws1; // open a new gap
tracer = 0;
}
n0 = rowV[j] + scaled_wg2;
if(rowF[j] >= n0) {
rowF[j] += scaled_ws2;
tracer |= kMaskFc;
}
else {
rowF[j] = n0 + scaled_ws2;
}
if (E >= rowF[j]) {
if(E >= G) {
V = E;
tracer |= kMaskE;
}
else {
V = G;
tracer |= kMaskD;
}
} else {
if(rowF[j] >= G) {
V = rowF[j];
}
else {
V = G;
tracer |= kMaskD;
}
}
backtrace_matrix.SetAt(k, tracer);
}
pV[j] = V;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
if(m_terminate = m_prg_callback(&m_prg_info)) {
break;
}
}
}
backtrace_matrix.Purge(k);
if(!m_terminate) {
x_DoBackTrace(backtrace_matrix, data);
}
return (TScore)(V + 0.5);
}
CNWAligner::TScore CMMAligner::x_RunTerm(const SCoordRect& rect,
bool left_top, bool right_bottom,
list<ETranscriptSymbol>& subpath)
{
if( m_terminate ) {
return 0;
}
const size_t N1 = rect.i2 - rect.i1 + 2;
const size_t N2 = rect.j2 - rect.j1 + 2;
vector<TScore> stl_rowV (N2), stl_rowF (N2);
TScore* rowV = &stl_rowV [0];
TScore* rowF = &stl_rowF [0];
// index calculation: [i,j] = i*n2 + j
vector<unsigned char> stl_bm (N1*N2);
unsigned char* backtrace = &stl_bm[0];
TScore* pV = rowV - 1;
const char* seq1 = m_Seq1 + rect.i1 - 1;
const char* seq2 = m_Seq2 + rect.j1 - 1;
const TNCBIScore (*sm) [NCBI_FSM_DIM] = m_ScoreMatrix.s;
bool bFreeGapLeft1 = m_esf_L1 && rect.i1 == 0;
bool bFreeGapRight1 = m_esf_R1 && rect.i2 == m_SeqLen1 - 1;
bool bFreeGapLeft2 = m_esf_L2 && rect.j1 == 0;
bool bFreeGapRight2 = m_esf_R2 && rect.j2 == m_SeqLen2 - 1;
TScore wgleft1 = bFreeGapLeft1? 0: m_Wg;
TScore wsleft1 = bFreeGapLeft1? 0: m_Ws;
TScore wg1 = m_Wg, ws1 = m_Ws;
// first row
size_t k;
{
rowV[0] = wgleft1;
for (k = 1; k < N2; k++) {
rowV[k] = pV[k] + wsleft1;
rowF[k] = kInfMinus;
backtrace[k] = kMaskE | kMaskEc;
}
rowV[0] = 0;
}
// recurrences
TScore wgleft2 = bFreeGapLeft2? 0: m_Wg;
TScore wsleft2 = bFreeGapLeft2? 0: m_Ws;
TScore V = 0;
TScore V0 = left_top? 0: wgleft2;
TScore E, G, n0;
unsigned char tracer;
size_t i, j;
for(i = 1; i < N1; ++i) {
V = V0 += wsleft2;
E = kInfMinus;
backtrace[k++] = kMaskFc;
unsigned char ci = seq1[i];
if(i == N1 - 1 && bFreeGapRight1) {
wg1 = ws1 = 0;
}
TScore wg2 = m_Wg, ws2 = m_Ws;
for (j = 1; j < N2; ++j, ++k) {
G = pV[j] + sm[ci][(unsigned char)seq2[j]];
pV[j] = V;
n0 = V + wg1;
if(E >= n0) {
E += ws1; // continue the gap
tracer = kMaskEc;
}
else {
E = n0 + ws1; // open a new gap
tracer = 0;
}
if(j == N2 - 1 && bFreeGapRight2) {
wg2 = ws2 = 0;
}
n0 = rowV[j] + ((right_bottom && j == N2 - 1)? 0: wg2);
if(rowF[j] >= n0) {
rowF[j] += ws2;
tracer |= kMaskFc;
}
else {
rowF[j] = n0 + ws2;
}
if (E >= rowF[j]) {
if(E >= G) {
V = E;
tracer |= kMaskE;
}
else {
V = G;
tracer |= kMaskD;
}
} else {
if(rowF[j] >= G) {
V = rowF[j];
}
else {
V = G;
tracer |= kMaskD;
}
}
backtrace[k] = tracer;
}
pV[j] = V;
}
// fill the subpath
subpath.clear();
// run backtrace
k = N1*N2 - 1;
while (k != 0) {
unsigned char Key = backtrace[k];
if (Key & kMaskD) {
subpath.push_front(eTS_Match);
k -= N2 + 1;
}
else if (Key & kMaskE) {
subpath.push_front(eTS_Insert); --k;
while(k > 0 && (Key & kMaskEc)) {
subpath.push_front(eTS_Insert);
Key = backtrace[k--];
}
}
else {
subpath.push_front(eTS_Delete);
k -= N2;
while(k > 0 && (Key & kMaskFc)) {
subpath.push_front(eTS_Delete);
Key = backtrace[k];
k -= N2;
}
}
}
return V;
}
void CMMAligner::x_RunBtm(const SCoordRect& rect,
vector<TScore>& vE, vector<TScore>& vF, vector<TScore>& vG,
vector<unsigned char>& trace, bool rb) const
{
if( m_terminate ) {
return;
}
const size_t dim1 = rect.i2 - rect.i1 + 1;
const size_t dim2 = rect.j2 - rect.j1 + 1;
const size_t N1 = dim1 + 1;
const size_t N2 = dim2 + 1;
vector<TScore> stl_rowV (N2), stl_rowF (N2);
TScore* rowV = &stl_rowV [0];
TScore* rowF = &stl_rowF [0];
TScore* pV = rowV + 1;
const char* seq1 = m_Seq1 + rect.i1;
const char* seq2 = m_Seq2 + rect.j1;
const TNCBIScore (*sm) [NCBI_FSM_DIM] = m_ScoreMatrix.s;
bool bFreeGapRight1 = m_esf_R1 && rect.i2 == m_SeqLen1 - 1;
bool bFreeGapRight2 = m_esf_R2 && rect.j2 == m_SeqLen2 - 1;
bool bFreeGapLeft2 = m_esf_L2 && rect.j1 == 0;
// progress reporting
const size_t prg_rep_rate = 100;
const size_t prg_rep_increment = prg_rep_rate*N2;
// bottom row
TScore wg = bFreeGapRight1? 0: m_Wg;
TScore ws = bFreeGapRight1? 0: m_Ws;
rowV[N2 - 1] = wg;
int i, j;
for (j = N2 - 2; j >= 0; --j) {
rowV[j] = pV[j] + ws;
rowF[j] = kInfMinus;
}
rowV[N2 - 1] = 0;
// recurrences
wg = bFreeGapRight2? 0: m_Wg;
ws = bFreeGapRight2? 0: m_Ws;
TScore V = 0;
TScore V0 = rb? 0: wg;
TScore E, G, n0;
for(i = N1 - 2; i > 0; --i) {
V = V0 += ws;
E = kInfMinus;
unsigned char ci = seq1[i];
TScore wg2 = m_Wg, ws2 = m_Ws;
for (j = N2 - 2; j >= 0; --j) {
G = pV[j] + sm[ci][(unsigned char)seq2[j]];
pV[j] = V;
n0 = V + m_Wg;
if(E >= n0)
E += m_Ws; // continue the gap
else
E = n0 + m_Ws; // open a new gap
if(j == 0 && bFreeGapLeft2) {
wg2 = ws2 = 0;
}
n0 = rowV[j] + wg2;
if (rowF[j] > n0)
rowF[j] += ws2;
else
rowF[j] = n0 + ws2;
V = (E >= rowF[j])? (E >= G? E: G): (rowF[j] >= G? rowF[j]: G);
}
pV[j] = V;
if( m_prg_callback && (N1 - i) % prg_rep_rate == 0 ) {
#ifdef NCBI_THREADS
CFastMutexGuard guard (progress_mutex);
#endif
m_prg_info.m_iter_done += prg_rep_increment;
if( (m_terminate = m_prg_callback(&m_prg_info)) ) {
break;
}
}
}
// the top row (i == 0)
if(!m_terminate) {
vF[N2-1] = V = V0 += ws;
vG[N2-1] = vE[N2-1] = E = kInfMinus;
trace[N2-1] = kMaskFc;
unsigned char ci = seq1[i];
TScore wg2 = m_Wg, ws2 = m_Ws;
unsigned char tracer;
for (j = N2 - 2; j >= 0; --j) {
vG[j] = G = pV[j] + sm[ci][(unsigned char)seq2[j]];
pV[j] = V;
n0 = V + m_Wg;
if(E >= n0) {
E += m_Ws; // continue the gap
tracer = kMaskEc;
}
else {
E = n0 + m_Ws; // open a new gap
tracer = 0;
}
vE[j] = E;
if(j == 0 && bFreeGapLeft2) {
wg2 = ws2 = 0;
}
n0 = rowV[j] + wg2;
if(rowF[j] >= n0) {
rowF[j] += ws2;
tracer |= kMaskFc;
}
else {
rowF[j] = n0 + ws2;
}
vF[j] = rowF[j];
if (E >= rowF[j]) {
if(E >= G) {
V = E;
tracer |= kMaskE;
}
else {
V = G;
tracer |= kMaskD;
}
} else {
if(rowF[j] >= G) {
V = rowF[j];
}
else {
V = G;
tracer |= kMaskD;
}
}
trace[j] = tracer;
}
}
if( m_prg_callback ) {
#ifdef NCBI_THREADS
CFastMutexGuard guard (progress_mutex);
#endif
m_prg_info.m_iter_done += (N1 - i) % prg_rep_rate;
m_terminate = m_prg_callback(&m_prg_info);
}
}
CNWAligner::TScore CNWAligner::x_Align(SAlignInOut* data)
{
//check data integrity
if( m_SmithWaterman && ( data->m_offset1 || m_SeqLen1 != data->m_len1 ||
data->m_offset2 || m_SeqLen2 != data->m_len2 ) ) {
NCBI_THROW(CAlgoAlignException, eBadParameter,
"Smith-Waterman not compatible with offsets provided");
}
if( m_SmithWaterman && ( !data->m_esf_L1 || !data->m_esf_R1 ||
!data->m_esf_L2 || !data->m_esf_R2 ) ) {
NCBI_THROW(CAlgoAlignException, eBadParameter,
"Smith-Waterman not compatible with end gap penalties");
}
const size_t N1 = data->m_len1 + 1;
const size_t N2 = data->m_len2 + 1;
vector<TScore> stl_rowV (N2), stl_rowF(N2);
const TNCBIScore (* sm) [NCBI_FSM_DIM] = m_ScoreMatrix.s;
if(m_prg_callback) {
m_prg_info.m_iter_total = N1*N2;
m_prg_info.m_iter_done = 0;
if( (m_terminate = m_prg_callback(&m_prg_info)) ) {
return 0;
}
}
bool bFreeGapLeft1 = data->m_esf_L1 && data->m_offset1 == 0;
bool bFreeGapRight1 = data->m_esf_R1 &&
m_SeqLen1 == data->m_offset1 + data->m_len1;
bool bFreeGapLeft2 = data->m_esf_L2 && data->m_offset2 == 0;
bool bFreeGapRight2 = data->m_esf_R2 &&
m_SeqLen2 == data->m_offset2 + data->m_len2;
TScore wgleft1 = bFreeGapLeft1? 0: m_Wg;
TScore wsleft1 = bFreeGapLeft1? 0: m_Ws;
TScore wg1 = m_Wg, ws1 = m_Ws;
// index calculation: [i,j] = i*n2 + j
CBacktraceMatrix4 backtrace_matrix (N1 * N2);
backtrace_matrix.SetAt(0, 0);
// first row
// note that stl_rowF[0] is not used in the main cycle,
size_t k;
stl_rowV[0] = wgleft1;
for (k = 1; k < N2; ++k) {
stl_rowV[k] = stl_rowV[k-1] + wsleft1;
stl_rowF[k] = kInfMinus;
backtrace_matrix.SetAt(k, kMaskE | kMaskEc);
}
backtrace_matrix.Purge(k);
stl_rowV[0] = 0;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
m_terminate = m_prg_callback(&m_prg_info);
}
// gap penalties
TScore wgleft2 (bFreeGapLeft2? 0: m_Wg);
TScore wsleft2 (bFreeGapLeft2? 0: m_Ws);
const char * seq1 = m_Seq1 + data->m_offset1;
const char * seq1_end = seq1 + data->m_len1;
TScore V0 = wgleft2;
TScore V = 0;//best score in the current cell. Will be equal to the NW score at the end
TScore best_V = 0;//best score in the whole matrix aka score for SW
--k;
for(; seq1 != seq1_end && !m_terminate; ++seq1) {
backtrace_matrix.SetAt(++k, kMaskFc);
if( seq1 + 1 == seq1_end && bFreeGapRight1) {
wg1 = ws1 = 0;
}
unsigned char tracer;
const TNCBIScore * row_sc = sm[(size_t)*seq1];
const char * seq2 = m_Seq2 + data->m_offset2;
const char * seq2_end = seq2 + data->m_len2;
TScore wg2 = m_Wg, ws2 = m_Ws;
//best ending with gap in seq1 open seq1 X- or extended seq1 X--
// seq2 XX seq2 XXX
TScore E = kInfMinus;
//best ending with gap in seq2
TScore F;
//total best with
//best ending with match
TScore G;
//just temporary
TScore n0;
//total best
TScore * rowV = &stl_rowV[0];//previos row
V = V0 += wsleft2; //current row
//best ending with match
TScore * rowF = &stl_rowF[0];
for (; seq2 != seq2_end;) {
G = *rowV + row_sc[(size_t)*seq2++];
*rowV = V;
n0 = V + wg1;
if(E >= n0) {
E += ws1; // continue the gap
tracer = kMaskEc;
}
else {
E = n0 + ws1; // open a new gap
tracer = 0;
}
if( bFreeGapRight2 && seq2 == seq2_end ) {
wg2 = ws2 = 0;
}
F = *++rowF;
n0 = *++rowV + wg2;
if(F >= n0) {
F += ws2;
tracer |= kMaskFc;
}
else {
F = n0 + ws2;
}
*rowF = F;
//best score
if( G < F || ( G == F && m_GapPreference == eLater) ) {
if( E <= F ) {
V = F;
} else {
V = E;
tracer |= kMaskE;
}
} else if( E > G || ( E == G && m_GapPreference == eLater) ) {
V = E;
tracer |= kMaskE;
} else {
V = G;
tracer |= kMaskD;
}
if (m_SmithWaterman && V < 0 ) {
V = 0;
}
backtrace_matrix.SetAt(++k, tracer);
if (V > best_V) {
best_V = V;
backtrace_matrix.SetBestPos(k);
}
}
*rowV = V;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
if( (m_terminate = m_prg_callback(&m_prg_info)) ) {
break;
}
}
}
backtrace_matrix.Purge(++k);
backtrace_matrix.SetBestScore(best_V);
/*
//print the matrix out
{{
cout<<endl;
int kk, ind1, ind2, width = 4;
cout<<setw(width)<<" ";
cout<<setw(width)<<"-";
for(ind2 = 0; ind2 < N2-1; ++ind2) {
cout<<setw(width)<<*(m_Seq2 + data->m_offset2 + ind2);
}
cout<<endl;
for(kk = 0,ind1 = 0; ind1 < N1; ++ind1) {
if(ind1) {
cout<<setw(width)<<(m_Seq1 + data->m_offset1)[ind1-1];
} else {
cout<<setw(width)<<"-";
}
for(ind2 = 0; ind2 < N2; ++ind2,++kk) {
string tstr;
unsigned char Key (backtrace_matrix[kk]);
if( Key & kMaskD ) tstr += "D";
else if ( Key & kMaskE ) tstr += "E";
else tstr += "F";
if( Key & kMaskEc ) tstr += "-";
if( Key & kMaskFc ) tstr += "|";
cout<<setw(width)<<tstr;
}
cout<<endl<<endl;
}
cout<<endl;
}}
//end of print the matrix out
*/
if(!m_terminate) {
x_SWDoBackTrace(backtrace_matrix, data);
//check back trace
TTranscript rv (data->m_transcript.size());
copy(data->m_transcript.rbegin(), data->m_transcript.rend(), rv.begin());
if(m_SmithWaterman) {
if( best_V != ScoreFromTranscript(rv, data->m_offset1, data->m_offset2) ) {
NCBI_THROW(CAlgoAlignException, eInternal,
"CNWAligner: error in back trace");
}
} else {
if( V != ScoreFromTranscript(rv, data->m_offset1, data->m_offset2) ) {
NCBI_THROW(CAlgoAlignException, eInternal,
"CNWAligner: error in back trace");
}
}
}
if(m_SmithWaterman) {
return best_V;
}
return V;
}
// Evaluate dynamic programming matrix. Create transcript.
CNWAligner::TScore CSplicedAligner32::x_Align (SAlignInOut* data)
{
// use the banded version if there is no space for introns
const int len_dif (data->m_len2 - data->m_len1);
if(len_dif < 2 * int (m_IntronMinSize) / 3) {
const Uint1 where (len_dif < 0? 0: 1);
const size_t shift (abs(len_dif) / 2);
const size_t band (abs(len_dif) + 2*(max(data->m_len1,data->m_len2)/20 + 1));
SetShift(where, shift);
SetBand(band);
return CBandAligner::x_Align(data);
}
// redefine TScore as a floating-point type for this procedure only
typedef double TScore;
const TScore cds_penalty_extra = -2e-6;
const size_t N1 = data->m_len1 + 1;
const size_t N2 = data->m_len2 + 1;
vector<TScore> stl_rowV (N2), stl_rowF (N2);
TScore* rowV = &stl_rowV[0];
TScore* rowF = &stl_rowF[0];
// index calculation: [i,j] = i*n2 + j
SAllocator<Uint4> alloc_bm (N1*N2);
Uint4* backtrace_matrix (alloc_bm.GetPointer());
TScore* pV = rowV - 1;
const char* seq1 = m_Seq1 + data->m_offset1 - 1;
const char* seq2 = m_Seq2 + data->m_offset2 - 1;
const TNCBIScore (*sm) [NCBI_FSM_DIM] = m_ScoreMatrix.s;
bool bFreeGapLeft1 = data->m_esf_L1 && data->m_offset1 == 0;
bool bFreeGapRight1 = data->m_esf_R1 &&
m_SeqLen1 == data->m_offset1 + data->m_len1;
bool bFreeGapLeft2 = data->m_esf_L2 && data->m_offset1 == 0;
bool bFreeGapRight2 = data->m_esf_R2 &&
m_SeqLen2 == data->m_offset2 + data->m_len2;
TScore wgleft1 = bFreeGapLeft1? 0: m_Wg;
TScore wsleft1 = bFreeGapLeft1? 0: m_Ws;
TScore wg1 = wgleft1, ws1 = wsleft1;
// recurrences
TScore wgleft2 = bFreeGapLeft2? 0: m_Wg;
TScore wsleft2 = bFreeGapLeft2? 0: m_Ws;
TScore V = 0;
TScore V0 = 0;
TScore E, G, n0;
Uint4 type;
// store candidate donors
size_t* jAllDonors [splice_type_count_32];
TScore* vAllDonors [splice_type_count_32];
vector<size_t> stl_jAllDonors (splice_type_count_32 * N2);
vector<TScore> stl_vAllDonors (splice_type_count_32 * N2);
for(unsigned char st = 0; st < splice_type_count_32; ++st) {
jAllDonors[st] = &stl_jAllDonors[st*N2];
vAllDonors[st] = &stl_vAllDonors[st*N2];
}
size_t jTail[splice_type_count_32], jHead[splice_type_count_32];
TScore vBestDonor [splice_type_count_32];
size_t jBestDonor [splice_type_count_32] = {0};
// place to store gap opening starts
size_t ins_start;
vector<size_t> stl_del_start(N2);
size_t* del_start = &stl_del_start[0];
// donor/acceptor matrix
const Uint1 * dnr_acc_matrix = g_dnr_acc_matrix.GetMatrix();
// fake row (above lambda)
rowV[0] = kInfMinus;
size_t k;
for (k = 0; k < N2; k++) {
rowV[k] = rowF[k] = kInfMinus;
del_start[k] = k;
}
k = 0;
size_t cds_start = m_cds_start, cds_stop = m_cds_stop;
if(cds_start < cds_stop) {
cds_start -= data->m_offset1;
cds_stop -= data->m_offset1;
}
size_t i, j = 0, k0;
unsigned char ci;
for(i = 0; i < N1; ++i, j = 0) {
V = i > 0? (V0 += wsleft2) : 0;
E = kInfMinus;
ins_start = k0 = k;
backtrace_matrix[k++] = kTypeGap; // | del_start[0]
ci = i > 0? seq1[i]: 'N';
for(unsigned char st = 0; st < splice_type_count_32; ++st) {
jTail[st] = jHead[st] = 0;
vBestDonor[st] = kInfMinus;
}
if(i == N1 - 1 && bFreeGapRight1) {
wg1 = ws1 = 0;
}
TScore wg2 = m_Wg, ws2 = m_Ws;
// detect donor candidate
if(N2 > 2) {
unsigned char d1 = seq2[1], d2 = seq2[2];
Uint1 dnr_type = 0xF0 & dnr_acc_matrix[(size_t(d1)<<8)|d2];
for(Uint1 st = 0; st < splice_type_count_32; ++st ) {
jAllDonors[st][jTail[st]] = j;
if(dnr_type & (0x10 << st)) {
vAllDonors[st][jTail[st]] =
( d1 == g_nwspl32_donor[st][0] &&
d2 == g_nwspl32_donor[st][1] ) ? V: (V + m_Wd1);
}
else { // both chars distorted
vAllDonors[st][jTail[st]] = V + m_Wd2;
}
++(jTail[st]);
}
}
if(cds_start <= i && i < cds_stop) {
if(i != 0 || ! bFreeGapLeft1) {
ws1 += cds_penalty_extra;
}
if(j != 0 || ! bFreeGapLeft2) {
ws2 += cds_penalty_extra;
}
}
for (j = 1; j < N2; ++j, ++k) {
G = pV[j] + sm[ci][(unsigned char)seq2[j]];
pV[j] = V;
n0 = V + wg1;
if(E >= n0) {
E += ws1; // continue the gap
}
else {
E = n0 + ws1; // open a new gap
ins_start = k-1;
}
if(j == N2 - 1 && bFreeGapRight2) {
wg2 = ws2 = 0;
}
n0 = rowV[j] + wg2;
if(rowF[j] >= n0) {
rowF[j] += ws2;
}
else {
rowF[j] = n0 + ws2;
del_start[j] = k-N2;
}
// evaluate the score (V)
if (E >= rowF[j]) {
if(E >= G) {
V = E;
type = kTypeGap | ins_start;
}
else {
V = G;
type = kTypeDiag;
}
} else {
if(rowF[j] >= G) {
V = rowF[j];
type = kTypeGap | del_start[j];
}
else {
V = G;
type = kTypeDiag;
}
}
// find out if there are new donors
for(unsigned char st = 0; st < splice_type_count_32; ++st) {
if(jTail[st] > jHead[st]) {
if(j - jAllDonors[st][jHead[st]] >= m_IntronMinSize) {
if(vAllDonors[st][jHead[st]] > vBestDonor[st]) {
vBestDonor[st] = vAllDonors[st][jHead[st]];
jBestDonor[st] = jAllDonors[st][jHead[st]];
}
++(jHead[st]);
}
}
}
// check splice signal
Uint4 dnr_pos = kMax_UI4;
unsigned char c1 = seq2[j-1], c2 = seq2[j];
Uint1 acc_mask = 0x0F & dnr_acc_matrix[(size_t(c1)<<8)|c2];
for(Uint1 st = 0; st < splice_type_count_32; ++st ) {
if(acc_mask & (0x01 << st)) {
TScore vAcc = vBestDonor[st] + m_Wi[st];
if( c1 != g_nwspl32_acceptor[st][0] ||
c2 != g_nwspl32_acceptor[st][1] ) {
vAcc += m_Wd1;
}
if(vAcc > V) {
V = vAcc;
dnr_pos = k0 + jBestDonor[st];
}
}
else { // try arbitrary splice
TScore vAcc = vBestDonor[st] + m_Wi[st] + m_Wd2;
if(vAcc > V) {
V = vAcc;
dnr_pos = k0 + jBestDonor[st];
}
}
}
if(dnr_pos != kMax_UI4) {
type = kTypeIntron | dnr_pos;
}
backtrace_matrix[k] = type;
// detect donor candidates
if(j < N2 - 2) {
unsigned char d1 = seq2[j+1], d2 = seq2[j+2];
Uint1 dnr_mask = 0xF0 & dnr_acc_matrix[(size_t(d1)<<8)|d2];
for(Uint1 st = 0; st < splice_type_count_32; ++st ) {
if( dnr_mask & (0x10 << st) ) {
if( d1 == g_nwspl32_donor[st][0] &&
d2 == g_nwspl32_donor[st][1] ) {
if(V > vBestDonor[st]) {
jAllDonors[st][jTail[st]] = j;
vAllDonors[st][jTail[st]] = V;
++(jTail[st]);
}
} else {
TScore v = V + m_Wd1;
if(v > vBestDonor[st]) {
jAllDonors[st][jTail[st]] = j;
vAllDonors[st][jTail[st]] = v;
++(jTail[st]);
}
}
}
else { // both chars distorted
TScore v = V + m_Wd2;
if(v > vBestDonor[st]) {
jAllDonors[st][jTail[st]] = j;
vAllDonors[st][jTail[st]] = v;
++(jTail[st]);
}
}
}
}
}
pV[j] = V;
if(i == 0) {
V0 = wgleft2;
wg1 = m_Wg;
ws1 = m_Ws;
}
}
try {
x_DoBackTrace(backtrace_matrix, data);
}
catch(exception&) { // GCC hack
throw;
}
return CNWAligner::TScore(V);
}
CNWAligner::TScore CNWAligner::x_Align(SAlignInOut* data)
{
const size_t N1 = data->m_len1 + 1;
const size_t N2 = data->m_len2 + 1;
vector<TScore> stl_rowV (N2), stl_rowF(N2);
TScore * rowV = &stl_rowV[0];
TScore * rowF = &stl_rowF[0];
TScore * pV = rowV - 1;
const char * seq1 = m_Seq1 + data->m_offset1 - 1;
const char * seq2 = m_Seq2 + data->m_offset2 - 1;
const TNCBIScore (* sm) [NCBI_FSM_DIM] = m_ScoreMatrix.s;
if(m_prg_callback) {
m_prg_info.m_iter_total = N1*N2;
m_prg_info.m_iter_done = 0;
if( (m_terminate = m_prg_callback(&m_prg_info)) ) {
return 0;
}
}
bool bFreeGapLeft1 = data->m_esf_L1 && data->m_offset1 == 0;
bool bFreeGapRight1 = data->m_esf_R1 &&
m_SeqLen1 == data->m_offset1 + data->m_len1;
bool bFreeGapLeft2 = data->m_esf_L2 && data->m_offset2 == 0;
bool bFreeGapRight2 = data->m_esf_R2 &&
m_SeqLen2 == data->m_offset2 + data->m_len2;
TScore wgleft1 = bFreeGapLeft1? 0: m_Wg;
TScore wsleft1 = bFreeGapLeft1? 0: m_Ws;
TScore wg1 = m_Wg, ws1 = m_Ws;
// index calculation: [i,j] = i*n2 + j
CBacktraceMatrix4 backtrace_matrix (N1 * N2);
backtrace_matrix.SetAt(0, 0);
// first row
size_t k;
rowV[0] = wgleft1;
for (k = 1; k < N2; ++k) {
rowV[k] = pV[k] + wsleft1;
rowF[k] = kInfMinus;
backtrace_matrix.SetAt(k, kMaskE | kMaskEc);
}
backtrace_matrix.Purge(k);
rowV[0] = 0;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
m_terminate = m_prg_callback(&m_prg_info);
}
// recurrences
TScore wgleft2 (bFreeGapLeft2? 0: m_Wg);
TScore wsleft2 (bFreeGapLeft2? 0: m_Ws);
TScore V (rowV[N2 - 1]);
TScore V0 (wgleft2);
TScore E, G, n0;
unsigned char tracer;
size_t i, j;
for(i = 1; i < N1 && !m_terminate; ++i) {
V = V0 += wsleft2;
E = kInfMinus;
backtrace_matrix.SetAt(k++, kMaskFc);
unsigned char ci = seq1[i];
if(i == N1 - 1 && bFreeGapRight1) {
wg1 = ws1 = 0;
}
TScore wg2 = m_Wg, ws2 = m_Ws;
for (j = 1; j < N2; ++j, ++k) {
G = pV[j] + sm[ci][(unsigned char)seq2[j]];
pV[j] = V;
n0 = V + wg1;
if(E >= n0) {
E += ws1; // continue the gap
tracer = kMaskEc;
}
else {
E = n0 + ws1; // open a new gap
tracer = 0;
}
if(j == N2 - 1 && bFreeGapRight2) {
wg2 = ws2 = 0;
}
n0 = rowV[j] + wg2;
if(rowF[j] >= n0) {
rowF[j] += ws2;
tracer |= kMaskFc;
}
else {
rowF[j] = n0 + ws2;
}
if (E >= rowF[j]) {
if(E >= G) {
V = E;
tracer |= kMaskE;
}
else {
V = G;
tracer |= kMaskD;
}
} else {
if(rowF[j] >= G) {
V = rowF[j];
}
else {
V = G;
tracer |= kMaskD;
}
}
backtrace_matrix.SetAt(k, tracer);
}
pV[j] = V;
if(m_prg_callback) {
m_prg_info.m_iter_done = k;
if( (m_terminate = m_prg_callback(&m_prg_info)) ) {
break;
}
}
}
backtrace_matrix.Purge(k);
if(!m_terminate) {
x_DoBackTrace(backtrace_matrix, data);
}
return V;
}
