CContextTranslator::CContextTranslator(const CSplitQueryBlk& sqb,
vector< CRef<IQueryFactory> >* query_chunk_factories /* = NULL */,
const CBlastOptions* options /* = NULL */)
{
const size_t kNumChunks(sqb.GetNumChunks());
m_ContextsPerChunk.reserve(kNumChunks);
for (size_t i = 0; i < kNumChunks; i++) {
m_ContextsPerChunk.push_back(sqb.GetQueryContexts(i));
}
if (query_chunk_factories == NULL || options == NULL) {
return;
}
/// Populate the data to print out
m_StartingChunks.resize(kNumChunks);
m_AbsoluteContexts.resize(kNumChunks);
for (size_t i = 0; i < kNumChunks; i++) {
CRef<IQueryFactory> chunk_qf((*query_chunk_factories)[i]);
CRef<ILocalQueryData> chunk_qd(chunk_qf->MakeLocalQueryData(options));
BlastQueryInfo* chunk_qinfo = chunk_qd->GetQueryInfo();
for (Int4 ctx = chunk_qinfo->first_context;
ctx <= chunk_qinfo->last_context; ctx++) {
m_StartingChunks[i].push_back(GetStartingChunk(i, ctx));
m_AbsoluteContexts[i].push_back(GetAbsoluteContext(i, ctx));
}
}
}
void
CQuerySplitter::x_ComputeContextOffsets_TranslatedQueries()
{
_ASSERT( !m_QueryChunkFactories.empty() );
const EBlastProgramType kProgram = m_Options->GetProgramType();
_ASSERT(Blast_QueryIsTranslated(kProgram));
const BlastQueryInfo* global_qinfo = m_LocalQueryData->GetQueryInfo();
#ifdef DEBUG_COMPARE_SEQUENCES
const BLAST_SequenceBlk* global_seq = m_LocalQueryData->GetSequenceBlk();
#endif
const size_t kOverlap =
SplitQuery_GetOverlapChunkSize(kProgram) / CODON_LENGTH;
CContextTranslator ctx_translator(*m_SplitBlk, &m_QueryChunkFactories,
m_Options);
CQueryDataPerChunk qdpc(*m_SplitBlk, kProgram, m_LocalQueryData);
vector<const BlastQueryInfo*> chunk_qinfo(m_NumChunks, 0);
for (size_t chunk_num = 0; chunk_num < m_NumChunks; chunk_num++) {
CRef<IQueryFactory> chunk_qf(m_QueryChunkFactories[chunk_num]);
CRef<ILocalQueryData> chunk_qd(chunk_qf->MakeLocalQueryData(m_Options));
#ifdef DEBUG_COMPARE_SEQUENCES
const BLAST_SequenceBlk* chunk_seq = chunk_qd->GetSequenceBlk();
#endif
// BlastQueryInfo structure corresponding to chunk number chunk_num
chunk_qinfo[chunk_num] = chunk_qd->GetQueryInfo();
_ASSERT(chunk_qinfo[chunk_num]);
// In case the first context differs from 0, for consistency with the
// other data returned by this class...
for (Int4 ctx = 0; ctx < chunk_qinfo[chunk_num]->first_context; ctx++) {
m_SplitBlk->AddContextOffsetToChunk(chunk_num, INT4_MAX);
}
for (Int4 ctx = chunk_qinfo[chunk_num]->first_context;
ctx <= chunk_qinfo[chunk_num]->last_context;
ctx++) {
size_t correction = 0;
const int starting_chunk =
ctx_translator.GetStartingChunk(chunk_num, ctx);
const int absolute_context =
ctx_translator.GetAbsoluteContext(chunk_num, ctx);
const int last_query_chunk = qdpc.GetLastChunk(chunk_num, ctx);
if (absolute_context == kInvalidContext ||
starting_chunk == kInvalidContext) {
_ASSERT( !chunk_qinfo[chunk_num]->contexts[ctx].is_valid );
// INT4_MAX is the sentinel value for invalid contexts
m_SplitBlk->AddContextOffsetToChunk(chunk_num, INT4_MAX);
continue;
}
// The corrections for the contexts corresponding to the negative
// strand in the last chunk of a query sequence are all 0
if (!s_IsPlusStrand(chunk_qinfo[chunk_num], ctx) &&
(chunk_num == (size_t)last_query_chunk) &&
(ctx % NUM_FRAMES >= 3)) {
correction = 0;
goto error_check;
}
// The corrections for the contexts corresponding to the plus
// strand are always the same, so only calculate the first one
if (s_IsPlusStrand(chunk_qinfo[chunk_num], ctx) &&
(ctx % NUM_FRAMES == 1 || ctx % NUM_FRAMES == 2)) {
correction = m_SplitBlk->GetContextOffsets(chunk_num).back();
goto error_check;
}
// If the query length is divisible by CODON_LENGTH, the
// corrections for all contexts corresponding to a given strand are
// the same, so only calculate the first one
if ((qdpc.GetQueryLength(chunk_num, ctx) % CODON_LENGTH == 0) &&
(ctx % NUM_FRAMES != 0) && (ctx % NUM_FRAMES != 3)) {
correction = m_SplitBlk->GetContextOffsets(chunk_num).back();
goto error_check;
}
// If the query length % CODON_LENGTH == 1, the corrections for the
// first two contexts of the negative strand are the same, and the
// correction for the last context is one more than that.
if ((qdpc.GetQueryLength(chunk_num, ctx) % CODON_LENGTH == 1) &&
!s_IsPlusStrand(chunk_qinfo[chunk_num], ctx)) {
if (ctx % NUM_FRAMES == 4) {
correction =
m_SplitBlk->GetContextOffsets(chunk_num).back();
goto error_check;
} else if (ctx % NUM_FRAMES == 5) {
correction =
m_SplitBlk->GetContextOffsets(chunk_num).back() + 1;
goto error_check;
}
}
// If the query length % CODON_LENGTH == 2, the corrections for the
// last two contexts of the negative strand are the same, which is
// one more that the first context on the negative strand.
if ((qdpc.GetQueryLength(chunk_num, ctx) % CODON_LENGTH == 2) &&
!s_IsPlusStrand(chunk_qinfo[chunk_num], ctx)) {
if (ctx % NUM_FRAMES == 4) {
correction =
m_SplitBlk->GetContextOffsets(chunk_num).back() + 1;
goto error_check;
} else if (ctx % NUM_FRAMES == 5) {
correction =
m_SplitBlk->GetContextOffsets(chunk_num).back();
goto error_check;
}
}
if (s_IsPlusStrand(chunk_qinfo[chunk_num], ctx)) {
for (int c = chunk_num; c != starting_chunk; c--) {
size_t prev_len = s_GetAbsoluteContextLength(chunk_qinfo,
c - 1,
ctx_translator,
absolute_context);
size_t curr_len = s_GetAbsoluteContextLength(chunk_qinfo, c,
ctx_translator,
absolute_context);
size_t overlap = min(kOverlap, curr_len);
correction += prev_len - min(overlap, prev_len);
}
} else {
size_t subtrahend = 0;
for (int c = chunk_num; c >= starting_chunk && c >= 0; c--) {
size_t prev_len = s_GetAbsoluteContextLength(chunk_qinfo,
c - 1,
ctx_translator,
absolute_context);
size_t curr_len = s_GetAbsoluteContextLength(chunk_qinfo,
c,
ctx_translator,
absolute_context);
size_t overlap = min(kOverlap, curr_len);
subtrahend += (curr_len - min(overlap, prev_len));
}
correction =
global_qinfo->contexts[absolute_context].query_length -
subtrahend;
}
error_check:
_ASSERT((chunk_qinfo[chunk_num]->contexts[ctx].is_valid));
m_SplitBlk->AddContextOffsetToChunk(chunk_num, correction);
#ifdef DEBUG_COMPARE_SEQUENCES
{
int global_offset = global_qinfo->contexts[absolute_context].query_offset +
correction;
int chunk_offset = chunk_qinfo[chunk_num]->contexts[ctx].query_offset;
int num_bases2compare =
min(10, chunk_qinfo[chunk_num]->contexts[ctx].query_length);
if (!cmp_sequence(&global_seq->sequence[global_offset],
&chunk_seq->sequence[chunk_offset],
num_bases2compare, Blast_QueryIsProtein(kProgram))) {
cerr << "Failed to compare sequence data for chunk " << chunk_num
<< ", context " << ctx << endl;
}
}
#endif
}
}
_TRACE("CContextTranslator contents: " << ctx_translator);
}
// Record the correction needed to synchronize with the complete query all the
// query offsets within HSPs that fall within this context
void
CQuerySplitter::x_ComputeContextOffsets_NonTranslatedQueries()
{
_ASSERT( !m_QueryChunkFactories.empty() );
const EBlastProgramType kProgram = m_Options->GetProgramType();
_ASSERT( !Blast_QueryIsTranslated(kProgram) );
const BlastQueryInfo* global_qinfo = m_LocalQueryData->GetQueryInfo();
#ifdef DEBUG_COMPARE_SEQUENCES
const BLAST_SequenceBlk* global_seq = m_LocalQueryData->GetSequenceBlk();
#endif
const size_t kOverlap = SplitQuery_GetOverlapChunkSize(kProgram);
CContextTranslator ctx_translator(*m_SplitBlk, &m_QueryChunkFactories,
m_Options);
vector<const BlastQueryInfo*> chunk_qinfo(m_NumChunks, 0);
for (size_t chunk_num = 0; chunk_num < m_NumChunks; chunk_num++) {
CRef<IQueryFactory> chunk_qf(m_QueryChunkFactories[chunk_num]);
CRef<ILocalQueryData> chunk_qd(chunk_qf->MakeLocalQueryData(m_Options));
#ifdef DEBUG_COMPARE_SEQUENCES
const BLAST_SequenceBlk* chunk_seq = chunk_qd->GetSequenceBlk();
#endif
// BlastQueryInfo structure corresponding to chunk number chunk_num
chunk_qinfo[chunk_num] = chunk_qd->GetQueryInfo();
_ASSERT(chunk_qinfo[chunk_num]);
// In case the first context differs from 0, for consistency with the
// other data returned by this class...
for (Int4 ctx = 0; ctx < chunk_qinfo[chunk_num]->first_context; ctx++) {
m_SplitBlk->AddContextOffsetToChunk(chunk_num, INT4_MAX);
}
for (Int4 ctx = chunk_qinfo[chunk_num]->first_context;
ctx <= chunk_qinfo[chunk_num]->last_context;
ctx++) {
size_t correction = 0;
const int starting_chunk =
ctx_translator.GetStartingChunk(chunk_num, ctx);
const int absolute_context =
ctx_translator.GetAbsoluteContext(chunk_num, ctx);
if (absolute_context == kInvalidContext ||
starting_chunk == kInvalidContext) {
_ASSERT( !chunk_qinfo[chunk_num]->contexts[ctx].is_valid );
// INT4_MAX is the sentinel value for invalid contexts
m_SplitBlk->AddContextOffsetToChunk(chunk_num, INT4_MAX);
continue;
}
if (s_IsPlusStrand(chunk_qinfo[chunk_num], ctx)) {
for (int c = chunk_num; c != starting_chunk; c--) {
size_t prev_len = s_GetAbsoluteContextLength(chunk_qinfo,
c - 1,
ctx_translator,
absolute_context);
size_t curr_len = s_GetAbsoluteContextLength(chunk_qinfo, c,
ctx_translator,
absolute_context);
size_t overlap = min(kOverlap, curr_len);
correction += prev_len - min(overlap, prev_len);
}
} else {
size_t subtrahend = 0;
for (int c = chunk_num; c >= starting_chunk && c >= 0; c--) {
size_t prev_len = s_GetAbsoluteContextLength(chunk_qinfo,
c - 1,
ctx_translator,
absolute_context);
size_t curr_len = s_GetAbsoluteContextLength(chunk_qinfo,
c,
ctx_translator,
absolute_context);
size_t overlap = min(kOverlap, curr_len);
subtrahend += (curr_len - min(overlap, prev_len));
}
correction =
global_qinfo->contexts[absolute_context].query_length -
subtrahend;
}
_ASSERT((chunk_qinfo[chunk_num]->contexts[ctx].is_valid));
m_SplitBlk->AddContextOffsetToChunk(chunk_num, correction);
#ifdef DEBUG_COMPARE_SEQUENCES
{
int global_offset = global_qinfo->contexts[absolute_context].query_offset +
correction;
int chunk_offset = chunk_qinfo[chunk_num]->contexts[ctx].query_offset;
if (!cmp_sequence(&global_seq->sequence[global_offset],
&chunk_seq->sequence[chunk_offset], 10,
Blast_QueryIsProtein(kProgram))) {
cerr << "Failed to compare sequence data!" << endl;
}
}
#endif
}
}
_TRACE("CContextTranslator contents: " << ctx_translator);
}
