WriteFlowDataClass::WriteFlowDataClass(unsigned int saveQueueSize, CommandLineOpts &inception_state, ImageSpecClass &my_image_spec, const RawWells & rawWells)
:queueSize(0),packQueuePtr(NULL),writeQueuePtr(NULL)
{
queueSize = saveQueueSize;
if(queueSize > 0){
size_t flowDepth = inception_state.bkg_control.signal_chunks.save_wells_flow;
size_t spaceSize = my_image_spec.rows * my_image_spec.cols;
packQueuePtr = new SemQueue();
writeQueuePtr = new SemQueue();
assert(packQueuePtr != NULL && writeQueuePtr != NULL);
packQueuePtr->init(queueSize);
writeQueuePtr->init(queueSize);
stepSize = rawWells.GetStepSize();
unsigned int bufferSize = stepSize * stepSize * flowDepth;
for(unsigned int item = 0; item < queueSize; ++item) {
ChunkFlowData* chunkData = new ChunkFlowData(spaceSize, flowDepth, bufferSize);
packQueuePtr->enQueue(chunkData);
}
filePath = rawWells.GetHdf5FilePath();
numCols = my_image_spec.cols;
saveAsUShort = inception_state.sys_context.well_convert;
}
}
void SetChipTypeFromWells(RawWells &rawWells)
{
if (rawWells.OpenMetaData()) //Use chip type stored in the wells file
{
if (rawWells.KeyExists("ChipType"))
{
string chipType;
rawWells.GetValue("ChipType", chipType);
ChipIdDecoder::SetGlobalChipId(chipType.c_str());
}
}
}
void output_debug_info(std::ostream &out, RawWells &wells) {
int rowStart, height, colStart, width;
wells.GetRegion(rowStart,height, colStart, width);
out << "w:" << rowStart << std::endl;
out << "w:" << height << std::endl;
out << "w:" << colStart << std::endl;
out << "w:" << width << std::endl;
}
void GetMetaDataForWells(char *dirExt, RawWells &rawWells, const char *chipType)
{
const char * paramsToKeep[] = {"Project","Sample","Start Time","Experiment Name","User Name","Serial Number","Oversample","Frame Time", "Num Frames", "Cycles", "Flows", "LibraryKeySequence", "ChipTemperature", "PGMTemperature", "PGMPressure","W2pH","W1pH","Cal Chip High/Low/InRange"};
std::string logFile = getExpLogPath(dirExt);
char* paramVal = NULL;
for (size_t pIx = 0; pIx < sizeof(paramsToKeep)/sizeof(char *); pIx++)
{
if ((paramVal = GetExpLogParameter(logFile.c_str(), paramsToKeep[pIx])) != NULL)
{
string value = paramVal;
size_t pos = value.find_last_not_of("\n\r \t");
if (pos != string::npos)
{
value = value.substr(0,pos+1);
}
rawWells.SetValue(paramsToKeep[pIx], value);
}
}
rawWells.SetValue("ChipType", chipType);
}
void SetWellsToLiveBeadsOnly(RawWells &rawWells, Mask *maskPtr)
{
// Get subset of wells we want to track, live only...
vector<int> subset;
size_t maskWells = maskPtr->H() * maskPtr->W();
subset.reserve(maskWells);
for (size_t i = 0; i < maskWells; i++) {
if (maskPtr->Match(i, MaskLive)) {
subset.push_back(i);
}
}
rawWells.SetSubsetToWrite(subset);
}
void CreateWellsFileForWriting (RawWells &rawWells, Mask *maskPtr,
CommandLineOpts &inception_state,
int numFlows,
int numRows, int numCols,
const char *chipType)
{
// set up wells data structure
MemUsage ("BeforeWells");
//rawWells.SetFlowChunkSize(flowChunk);
rawWells.SetCompression (inception_state.bkg_control.signal_chunks.wellsCompression);
rawWells.SetRows (numRows);
rawWells.SetCols (numCols);
rawWells.SetFlows (numFlows);
rawWells.SetFlowOrder (inception_state.flow_context.flowOrder); // 6th duplicated code
SetWellsToLiveBeadsOnly (rawWells,maskPtr);
// any model outputs a wells file of this nature
GetMetaDataForWells ((char*)(inception_state.sys_context.explog_path.c_str()),rawWells,chipType);
rawWells.OpenForWrite();
rawWells.WriteRanks(); // dummy, written for completeness
rawWells.WriteInfo(); // metadata written, do not need to rewrite
rawWells.Close(); // just create in this routine
MemUsage ("AfterWells");
}
void count_sample(filter_counts& counts, deque<float>& ppf, deque<float>& ssq, Mask& mask, RawWells& wells, const vector<int>& key_ionogram)
{
// Take sample of reads from a RawWells file, and apply some simple
// filters to identify problem reads.
// Record number of reads in sample, and number of reads caught by
// each filter.
well_set sample = sample_lib(mask, counts._nsamp);
WellData data;
unsigned int nflows = wells.NumFlows();
vector<float> nrm(nflows);
int flow0 = mixed_first_flow();
int flow1 = mixed_last_flow();
wells.ResetCurrentRegionWell();
// Some temporary code for comparing clonal filter in background model:
ofstream out("basecaller_ppf_ssq.txt");
assert(out);
while(!wells.ReadNextRegionData(&data)){
// Skip if this is not in the sample:
well_coord wc(data.y, data.x);
if(sample.find(wc) == sample.end())
continue;
// Skip wells with infinite signal:
bool finite = all_finite(data.flowValues, data.flowValues+nflows);
if(not finite){
++counts._ninf;
continue;
}
// Key-normalize:
float normalizer = ComputeNormalizerKeyFlows(data.flowValues, &key_ionogram[0], key_ionogram.size());
transform(data.flowValues, data.flowValues+nflows, nrm.begin(), bind2nd(divides<float>(),normalizer));
// Skip wells with bad key:
bool good_key = key_is_good(nrm.begin(), key_ionogram.begin(), key_ionogram.end());
if(not good_key){
++counts._nbad_key;
continue;
}
// Skip possible super-mixed beads:
float perc_pos = percent_positive(nrm.begin()+flow0, nrm.begin()+flow1);;
if(perc_pos > mixed_ppf_cutoff()){
++counts._nsuper;
continue;
}
// Record ppf and ssq:
float sum_frac = sum_fractional_part(nrm.begin()+flow0, nrm.begin()+flow1);
ppf.push_back(perc_pos);
ssq.push_back(sum_frac);
// Some temporary code for comparing clonal filter in background model:
out << setw(6) << data.y
<< setw(6) << data.x
<< setw(8) << setprecision(2) << fixed << perc_pos
<< setw(8) << setprecision(2) << fixed << sum_frac
<< setw(8) << setprecision(2) << fixed << normalizer
<< endl;
}
assert(ppf.size() == ssq.size());
}
void * BasecallerWorker(void *input)
{
BaseCallerContext& bc = *static_cast<BaseCallerContext*>(input);
RawWells wells ("", bc.filename_wells.c_str());
pthread_mutex_lock(&bc.mutex);
wells.OpenForIncrementalRead();
pthread_mutex_unlock(&bc.mutex);
vector<float> residual(bc.flow_order.num_flows(), 0);
vector<float> scaled_residual(bc.flow_order.num_flows(), 0);
vector<float> wells_measurements(bc.flow_order.num_flows(), 0);
vector<float> local_noise(bc.flow_order.num_flows(), 0);
vector<float> minus_noise_overlap(bc.flow_order.num_flows(), 0);
vector<float> homopolymer_rank(bc.flow_order.num_flows(), 0);
vector<float> neighborhood_noise(bc.flow_order.num_flows(), 0);
vector<float> phasing_parameters(3);
vector<uint16_t> flowgram(bc.flow_order.num_flows());
vector<int16_t> flowgram2(bc.flow_order.num_flows());
vector<int16_t> filtering_details(13,0);
vector<char> abParams;
abParams.reserve(256);
vector<uint8_t> quality(3*bc.flow_order.num_flows());
vector<int> base_to_flow (3*bc.flow_order.num_flows()); //!< Flow of in-phase incorporation of each base.
TreephaserSSE treephaser_sse(bc.flow_order, bc.windowSize);
DPTreephaser treephaser(bc.flow_order, bc.windowSize);
treephaser.SetStateProgression(bc.diagonal_state_prog);
treephaser.SkipRecalDuringNormalization(bc.skip_recal_during_norm);
treephaser_sse.SkipRecalDuringNormalization(bc.skip_recal_during_norm);
while (true) {
//
// Step 1. Retrieve next unprocessed region
//
pthread_mutex_lock(&bc.mutex);
int current_region, begin_x, begin_y, end_x, end_y;
if (not bc.chip_subset.GetCurrentRegionAndIncrement(current_region, begin_x, end_x, begin_y, end_y)) {
wells.Close();
pthread_mutex_unlock(&bc.mutex);
return NULL;
}
int num_usable_wells = 0;
for (int y = begin_y; y < end_y; ++y)
for (int x = begin_x; x < end_x; ++x)
if (bc.class_map[x + y * bc.chip_subset.GetChipSizeX()] >= 0)
num_usable_wells++;
if (begin_x == 0) printf("\n% 5d/% 5d: ", begin_y, bc.chip_subset.GetChipSizeY());
if (num_usable_wells == 0) printf(" ");
else if (num_usable_wells < 750) printf(". ");
else if (num_usable_wells < 1500) printf("o ");
else if (num_usable_wells < 2250) printf("# ");
else printf("##");
fflush(NULL);
if (begin_x == 0)
SaveBaseCallerProgress(10 + (80*begin_y)/bc.chip_subset.GetChipSizeY(), bc.output_directory);
pthread_mutex_unlock(&bc.mutex);
// Process the data
deque<ProcessedRead> lib_reads; // Collection of template library reads
deque<ProcessedRead> tf_reads; // Collection of test fragment reads
deque<ProcessedRead> calib_reads; // Collection of calibration library reads
deque<ProcessedRead> unfiltered_reads; // Random subset of lib_reads
deque<ProcessedRead> unfiltered_trimmed_reads; // Random subset of lib_reads
if (num_usable_wells == 0) { // There is nothing in this region. Don't even bother reading it
bc.lib_writer.WriteRegion(current_region, lib_reads);
if (bc.have_calibration_panel)
bc.calib_writer.WriteRegion(current_region, calib_reads);
if (bc.process_tfs)
bc.tf_writer.WriteRegion(current_region, tf_reads);
if (!bc.unfiltered_set.empty()) {
bc.unfiltered_writer.WriteRegion(current_region,unfiltered_reads);
bc.unfiltered_trimmed_writer.WriteRegion(current_region,unfiltered_trimmed_reads);
}
continue;
}
wells.SetChunk(begin_y, end_y-begin_y, begin_x, end_x-begin_x, 0, bc.flow_order.num_flows());
wells.ReadWells();
for (int y = begin_y; y < end_y; ++y)
for (int x = begin_x; x < end_x; ++x) { // Loop over wells within current region
//
// Step 2. Retrieve additional information needed to process this read
//
unsigned int read_index = x + y * bc.chip_subset.GetChipSizeX();
int read_class = bc.class_map[read_index];
if (read_class < 0)
continue;
bool is_random_calibration_read = false;
if (read_class == 2){
is_random_calibration_read = true;
read_class = 0; // Calibration reads are library beads;
}
bool is_random_unfiltered = bc.unfiltered_set.count(read_index) > 0;
if (not is_random_unfiltered and bc.only_process_unfiltered_set)
continue;
bc.filters->SetValid(read_index); // Presume valid until some filter proves otherwise
if (read_class == 0)
lib_reads.push_back(ProcessedRead(bc.barcodes->NoBarcodeReadGroup()));
else
tf_reads.push_back(ProcessedRead(0));
ProcessedRead& processed_read = (read_class==0) ? lib_reads.back() : tf_reads.back();
// Respect filter decisions from Background Model
if (bc.mask->Match(read_index, MaskFilteredBadResidual))
bc.filters->SetBkgmodelHighPPF(read_index, processed_read.filter);
if (bc.mask->Match(read_index, MaskFilteredBadPPF))
bc.filters->SetBkgmodelPolyclonal(read_index, processed_read.filter);
if (bc.mask->Match(read_index, MaskFilteredBadKey))
bc.filters->SetBkgmodelFailedKeypass(read_index, processed_read.filter);
if (!is_random_unfiltered and !bc.filters->IsValid(read_index)) // No reason to waste more time
continue;
float cf = bc.estimator.GetWellCF(x,y);
float ie = bc.estimator.GetWellIE(x,y);
float dr = bc.estimator.GetWellDR(x,y);
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow)
wells_measurements[flow] = wells.At(y,x,flow);
// Sanity check. If there are NaNs in this read, print warning
vector<int> nanflow;
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow) {
if (!isnan(wells_measurements[flow]))
continue;
wells_measurements[flow] = 0;
nanflow.push_back(flow);
}
if (nanflow.size() > 0) {
fprintf(stderr, "ERROR: BaseCaller read NaNs from wells file, x=%d y=%d flow=%d", x, y, nanflow[0]);
for (unsigned int flow=1; flow < nanflow.size(); flow++) {
fprintf(stderr, ",%d", nanflow[flow]);
}
fprintf(stderr, "\n");
fflush(stderr);
}
//
// Step 3. Perform base calling and quality value calculation
//
BasecallerRead read;
bool key_pass = true;
if (bc.keynormalizer == "keynorm-new") {
key_pass = read.SetDataAndKeyNormalizeNew(&wells_measurements[0], wells_measurements.size(), bc.keys[read_class].flows(), bc.keys[read_class].flows_length() - 1, false);
} else { // if (bc.keynormalizer == "keynorm-old") {
key_pass = read.SetDataAndKeyNormalize(&wells_measurements[0], wells_measurements.size(), bc.keys[read_class].flows(), bc.keys[read_class].flows_length() - 1);
}
// Get rid of outliers quickly
bc.filters->FilterHighPPFAndPolyclonal (read_index, read_class, processed_read.filter, read.raw_measurements, bc.polyclonal_filter);
if (not key_pass)
bc.filters->FilterFailedKeypass (read_index, read_class, processed_read.filter, read.sequence);
if (!is_random_unfiltered and !bc.filters->IsValid(read_index)) // No reason to waste more time
continue;
// Check if this read is either from the calibration panel or from the random calibration set
if(bc.calibration_training and bc.have_calibration_panel) {
if (!is_random_calibration_read and !bc.calibration_barcodes->MatchesBarcodeSignal(read)) {
bc.filters->SetFiltered(read_index, read_class, processed_read.filter); // Set as filtered
continue; // And move on along
}
}
// Equal recalibration opportunity for everybody! (except TFs!)
const vector<vector<vector<float> > > * aPtr = 0;
const vector<vector<vector<float> > > * bPtr = 0;
if (bc.recalModel.is_enabled() && read_class == 0) { //do not recalibrate TF read bc.chip_subset.GetChipSizeX()
aPtr = bc.recalModel.getAs(x+bc.chip_subset.GetColOffset(), y+bc.chip_subset.GetRowOffset());
bPtr = bc.recalModel.getBs(x+bc.chip_subset.GetColOffset(), y+bc.chip_subset.GetRowOffset());
}
// Execute the iterative solving-normalization routine - switch by specified algorithm
if (bc.dephaser == "treephaser-sse") {
treephaser_sse.SetAsBs(aPtr, bPtr); // Set/delete recalibration model for this read
treephaser_sse.SetModelParameters(cf, ie); // sse version has no hookup for droop.
treephaser_sse.NormalizeAndSolve(read);
treephaser.SetModelParameters(cf, ie); // Adapter trimming uses the cpp treephaser
} else { // Setup cpp treephaser
if (bc.skip_droop)
treephaser.SetModelParameters(cf, ie);
else
treephaser.SetModelParameters(cf, ie, dr);
treephaser.SetAsBs(aPtr, bPtr); // Set/delete recalibration model for this read
if (bc.dephaser == "dp-treephaser") {
// Single parameter gain estimation
treephaser.NormalizeAndSolve_GainNorm(read, bc.flow_order.num_flows());
} else if (bc.dephaser == "treephaser-adaptive") {
// Adaptive nortmalization - resolving read from start in each iteration
treephaser.NormalizeAndSolve_Adaptive(read, bc.flow_order.num_flows());
} else { //if (bc.dephaser == "treephaser-swan") {
// Default corresponding to (approximately) what the sse version is doing
// Adaptive normalization - sliding window without resolving start
treephaser.NormalizeAndSolve_SWnorm(read, bc.flow_order.num_flows());
}
// Need this function to calculate inphase population for cpp version
treephaser.ComputeQVmetrics(read);
}
// If recalibration is enabled, generate adjusted sequence and normalized_measurements, and recompute QV metrics
bool calibrate_read = (bc.recalibration.is_enabled() && read_class == 0); //do not recalibrate TF read
if (calibrate_read) {
// Change base sequence for low hps
bc.recalibration.CalibrateRead(x+bc.chip_subset.GetColOffset(),y+bc.chip_subset.GetRowOffset(),read.sequence, read.normalized_measurements, read.prediction, read.state_inphase);
if (bc.dephaser == "treephaser-sse")
treephaser_sse.ComputeQVmetrics(read);
else
treephaser.ComputeQVmetrics(read);
} else if (bc.dephaser == "treephaser-sse") {
// in case we didn't calibrate low hps, still want to have QV metrics for sse output
treephaser_sse.ComputeQVmetrics(read);
}
// Misc data management: Generate residual, scaled_residual
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow) {
residual[flow] = read.normalized_measurements[flow] - read.prediction[flow];
scaled_residual[flow] = residual[flow] / read.state_inphase[flow];
}
// Misc data management: Put base calls in proper string form
processed_read.filter.n_bases = read.sequence.size();
processed_read.filter.is_called = true;
// Misc data management: Generate base_to_flow
base_to_flow.clear();
base_to_flow.reserve(processed_read.filter.n_bases);
for (int base = 0, flow = 0; base < processed_read.filter.n_bases; ++base) {
while (flow < bc.flow_order.num_flows() and read.sequence[base] != bc.flow_order[flow])
flow++;
base_to_flow.push_back(flow);
}
// Misc data management: Populate some trivial read properties
char read_name[256];
sprintf(read_name, "%s:%05d:%05d", bc.run_id.c_str(), bc.chip_subset.GetRowOffset() + y, bc.chip_subset.GetColOffset() + x);
processed_read.bam.Name = read_name;
processed_read.bam.SetIsMapped(false);
phasing_parameters[0] = cf;
phasing_parameters[1] = ie;
phasing_parameters[2] = dr;
processed_read.bam.AddTag("ZP", phasing_parameters);
// Calculation of quality values
// Predictor 1 - Treephaser residual penalty
// Predictor 2 - Local noise/flowalign - 'noise' in the input base's measured val. Noise is max[abs(val - round(val))] within +-1 BASES
// Predictor 3 - Read Noise/Overlap - mean & stdev of the 0-mers & 1-mers in the read
// Predictor 3 (new) - Beverly Events
// Predictor 4 - Transformed homopolymer length
// Predictor 5 - Treephaser: Penalty indicating deletion after the called base
// Predictor 6 - Neighborhood noise - mean of 'noise' +-5 BASES around a base. Noise is mean{abs(val - round(val))}
int num_predictor_bases = min(bc.flow_order.num_flows(), processed_read.filter.n_bases);
PerBaseQual::PredictorLocalNoise(local_noise, num_predictor_bases, base_to_flow, read.normalized_measurements, read.prediction);
PerBaseQual::PredictorNeighborhoodNoise(neighborhood_noise, num_predictor_bases, base_to_flow, read.normalized_measurements, read.prediction);
//PerBaseQual::PredictorNoiseOverlap(minus_noise_overlap, num_predictor_bases, read.normalized_measurements, read.prediction);
PerBaseQual::PredictorBeverlyEvents(minus_noise_overlap, num_predictor_bases, base_to_flow, scaled_residual);
PerBaseQual::PredictorHomopolymerRank(homopolymer_rank, num_predictor_bases, read.sequence);
quality.clear();
bc.quality_generator.GenerateBaseQualities(processed_read.bam.Name, processed_read.filter.n_bases, bc.flow_order.num_flows(),
read.penalty_residual, local_noise, minus_noise_overlap, // <- predictors 1,2,3
homopolymer_rank, read.penalty_mismatch, neighborhood_noise, // <- predictors 4,5,6
base_to_flow, quality,
read.additive_correction,
read.multiplicative_correction,
read.state_inphase);
//
// Step 4a. Barcode classification of library reads
//
if (processed_read.filter.n_bases_filtered == -1)
processed_read.filter.n_bases_filtered = processed_read.filter.n_bases;
processed_read.filter.n_bases_key = min(bc.keys[read_class].bases_length(), processed_read.filter.n_bases);
processed_read.filter.n_bases_prefix = processed_read.filter.n_bases_key;
processed_read.barcode_n_errors = 0;
if (read_class == 0)
{ // Library beads - first separate out calibration barcodes
processed_read.read_group_index = -1;
if (bc.have_calibration_panel){
bc.calibration_barcodes->ClassifyAndTrimBarcode(read_index, processed_read, read, base_to_flow);
processed_read.is_control_barcode = (processed_read.read_group_index >= 0);
}
if (processed_read.read_group_index < 0)
bc.barcodes->ClassifyAndTrimBarcode(read_index, processed_read, read, base_to_flow);
}
//
// Step 4b. Custom mod: Trim extra bases after key and barcode. Make it look like barcode trimming.
//
if (bc.extra_trim_left > 0)
processed_read.filter.n_bases_prefix = min(processed_read.filter.n_bases_prefix + bc.extra_trim_left, processed_read.filter.n_bases);
//
// Step 4. Calculate/save read metrics and apply filters
//
bc.filters->FilterZeroBases (read_index, read_class, processed_read.filter);
bc.filters->FilterShortRead (read_index, read_class, processed_read.filter);
bc.filters->FilterFailedKeypass (read_index, read_class, processed_read.filter, read.sequence);
bc.filters->FilterHighResidual (read_index, read_class, processed_read.filter, residual);
bc.filters->FilterBeverly (read_index, read_class, processed_read.filter, scaled_residual, base_to_flow);
bc.filters->FilterQuality (read_index, read_class, processed_read.filter, quality);
bc.filters->TrimAdapter (read_index, read_class, processed_read, scaled_residual, base_to_flow, treephaser, read);
bc.filters->TrimQuality (read_index, read_class, processed_read.filter, quality);
bc.filters->TrimAvalanche (read_index, read_class, processed_read.filter, quality);
//! New mechanism for dumping potentially useful metrics.
if (bc.metric_saver->save_anything() and (is_random_unfiltered or !bc.metric_saver->save_subset_only())) {
pthread_mutex_lock(&bc.mutex);
bc.metric_saver->SaveRawMeasurements (y,x,read.raw_measurements);
bc.metric_saver->SaveAdditiveCorrection (y,x,read.additive_correction);
bc.metric_saver->SaveMultiplicativeCorrection (y,x,read.multiplicative_correction);
bc.metric_saver->SaveNormalizedMeasurements (y,x,read.normalized_measurements);
bc.metric_saver->SavePrediction (y,x,read.prediction);
bc.metric_saver->SaveStateInphase (y,x,read.state_inphase);
bc.metric_saver->SaveStateTotal (y,x,read.state_total);
bc.metric_saver->SavePenaltyResidual (y,x,read.penalty_residual);
bc.metric_saver->SavePenaltyMismatch (y,x,read.penalty_mismatch);
bc.metric_saver->SaveLocalNoise (y,x,local_noise);
bc.metric_saver->SaveNoiseOverlap (y,x,minus_noise_overlap);
bc.metric_saver->SaveHomopolymerRank (y,x,homopolymer_rank);
bc.metric_saver->SaveNeighborhoodNoise (y,x,neighborhood_noise);
pthread_mutex_unlock(&bc.mutex);
}
//
// Step 4b. Add flow signal information to ZM tag in BAM record.
//
flowgram2.clear();
int max_flow = min(bc.flow_order.num_flows(),16);
if (processed_read.filter.n_bases_filtered > 0)
max_flow = min(bc.flow_order.num_flows(), base_to_flow[processed_read.filter.n_bases_filtered-1] + 16);
vector<int> out_of_boud_flows;
for (int flow = 0; flow < max_flow; ++flow){
float temp_flowgram = 128*read.normalized_measurements[flow];
if (temp_flowgram < -16383.0f or temp_flowgram > 16383.0f) {
out_of_boud_flows.push_back(flow);
temp_flowgram = min(max(-16383.0f,temp_flowgram), 16383.0f);
}
//flowgram2.push_back(2*(int16_t)(128*read.normalized_measurements[flow]));
flowgram2.push_back(2*(int16_t)temp_flowgram);
}
// Do not spam stderr
/*if (out_of_boud_flows.size() > 0) {
cerr << "BaseCaller WARNING: Normalized signal out of bounds in well y="
<< y << ", x=" << x << ", in flows ";
for (unsigned int flow = 0; flow < out_of_boud_flows.size()-1; ++flow)
cerr << out_of_boud_flows.at(flow) << ',';
cerr << out_of_boud_flows.at(out_of_boud_flows.size()-1) << endl;
} */
processed_read.bam.AddTag("ZM", flowgram2);
//flowgram2.push_back(1*(int16_t)(256*read.normalized_measurements[flow]));
//flowgram2.push_back(2*(int16_t)(128*read.normalized_measurements[flow]));
//flowgram2.push_back(4*(int16_t)(64*read.normalized_measurements[flow]));
//flowgram2.push_back(8*(int16_t)(32*read.normalized_measurements[flow]));
//
// Step 4c. Populate FZ tag in BAM record.
//
flowgram.clear();
if (bc.flow_signals_type == "wells") {
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow)
flowgram.push_back(max(0,(int)(100.0*wells_measurements[flow]+0.5)));
processed_read.bam.AddTag("FZ", flowgram); // Will be phased out soon
} else if (bc.flow_signals_type == "key-normalized") {
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow)
flowgram.push_back(max(0,(int)(100.0*read.raw_measurements[flow]+0.5)));
processed_read.bam.AddTag("FZ", flowgram); // Will be phased out soon
} else if (bc.flow_signals_type == "adaptive-normalized") {
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow)
flowgram.push_back(max(0,(int)(100.0*read.normalized_measurements[flow]+0.5)));
processed_read.bam.AddTag("FZ", flowgram); // Will be phased out soon
} else if (bc.flow_signals_type == "residual") {
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow)
flowgram.push_back(max(0,(int)(1000 + 100*residual[flow])));
processed_read.bam.AddTag("FZ", flowgram); // Will be phased out soon
} else if (bc.flow_signals_type == "scaled-residual") { // This settings is necessary part of calibration training
for (int flow = 0; flow < bc.flow_order.num_flows(); ++flow) {
//between 0 and 98
float adjustment = min(0.49f, max(-0.49f, scaled_residual[flow]));
flowgram.push_back(max(0,(int)(49.5 + 100*adjustment)));
}
processed_read.bam.AddTag("FZ", flowgram);
}
//
// Step 5. Pass basecalled reads to appropriate writers
//
// Create BAM entries
if (processed_read.filter.n_bases > 0) {
processed_read.bam.QueryBases.reserve(processed_read.filter.n_bases);
processed_read.bam.Qualities.reserve(processed_read.filter.n_bases);
for (int base = processed_read.filter.n_bases_prefix; base < processed_read.filter.n_bases_filtered; ++base) {
processed_read.bam.QueryBases.push_back(read.sequence[base]);
processed_read.bam.Qualities.push_back(quality[base] + 33);
}
processed_read.bam.AddTag("ZF","i", base_to_flow[processed_read.filter.n_bases_prefix]);
} else
processed_read.bam.AddTag("ZF","i", 0);
// Randomly selected library beads - excluding calibration reads
if (is_random_unfiltered and (not processed_read.is_control_barcode)) {
unfiltered_trimmed_reads.push_back(processed_read);
unfiltered_reads.push_back(processed_read);
ProcessedRead& untrimmed_read = unfiltered_reads.back();
processed_read.filter.GenerateZDVector(filtering_details);
untrimmed_read.bam.AddTag("ZD", filtering_details);
if (processed_read.filter.n_bases > 0) {
untrimmed_read.bam.QueryBases.reserve(processed_read.filter.n_bases);
untrimmed_read.bam.Qualities.reserve(processed_read.filter.n_bases);
for (int base = max(processed_read.filter.n_bases_filtered,processed_read.filter.n_bases_prefix); base < processed_read.filter.n_bases; ++base) {
untrimmed_read.bam.QueryBases.push_back(read.sequence[base]);
untrimmed_read.bam.Qualities.push_back(quality[base] + 33);
}
}
// Temporary workaround: provide fake FZ tag for unfiltered.trimmed and unfiltered.untrimmed sets.
if (bc.flow_signals_type == "none") {
flowgram.assign(1,0);
unfiltered_reads.back().bam.AddTag("FZ", flowgram);
unfiltered_trimmed_reads.back().bam.AddTag("FZ", flowgram);
}
// If this read was supposed to have "early filtering", make sure we emulate that here
if (processed_read.filter.n_bases_after_bkgmodel_bad_key >= 0 or
processed_read.filter.n_bases_after_bkgmodel_high_ppf >= 0 or
processed_read.filter.n_bases_after_bkgmodel_polyclonal >= 0 or
processed_read.filter.n_bases_after_high_ppf >= 0 or
processed_read.filter.n_bases_after_polyclonal >= 0)
processed_read.filter.n_bases = -1;
}
// Move read from lib_reads stack to calib_reads if necessary
// This invalidates the processed_read reference and needs to be at the very end
if (processed_read.is_control_barcode) {
calib_reads.push_back(processed_read);
lib_reads.pop_back();
}
}
bc.lib_writer.WriteRegion(current_region, lib_reads);
if (bc.have_calibration_panel)
bc.calib_writer.WriteRegion(current_region, calib_reads);
if (bc.process_tfs)
bc.tf_writer.WriteRegion(current_region, tf_reads);
if (!bc.unfiltered_set.empty()) {
bc.unfiltered_writer.WriteRegion(current_region,unfiltered_reads);
bc.unfiltered_trimmed_writer.WriteRegion(current_region,unfiltered_trimmed_reads);
}
}
}
int main (int argc, const char *argv[])
{
BaseCallerSalute();
time_t analysis_start_time;
time(&analysis_start_time);
Json::Value basecaller_json(Json::objectValue);
DumpStartingStateOfProgram (argc,argv,analysis_start_time, basecaller_json["BaseCaller"]);
//
// Step 1. Process Command Line Options & Initialize Modules
//
BaseCallerParameters bc_params;
OptArgs opts, null_opts;
opts.ParseCmdLine(argc, argv);
if (opts.GetFirstBoolean('h', "help", false) or argc == 1)
bc_params.PrintHelp();
if (opts.GetFirstBoolean('v', "version", false)) {
fprintf (stdout, "%s", IonVersion::GetFullVersion ("BaseCaller").c_str());
exit (EXIT_SUCCESS);
}
// Command line processing *** Main directories and file locations first
bc_params.InitializeFilesFromOptArgs(opts);
bc_params.InitContextVarsFromOptArgs(opts);
// Command line processing *** Options that have default values retrieved from wells or mask files
RawWells wells ("", bc_params.GetFiles().filename_wells.c_str());
if (!wells.OpenMetaData()) {
fprintf (stderr, "Failed to retrieve metadata from %s\n", bc_params.GetFiles().filename_wells.c_str());
exit (EXIT_FAILURE);
}
Mask mask (1, 1);
if (mask.SetMask (bc_params.GetFiles().filename_mask.c_str()))
exit (EXIT_FAILURE);
string chip_type = "unknown";
if (wells.KeyExists("ChipType"))
wells.GetValue("ChipType", chip_type);
// Command line processing *** Various general option and opts to classify and sample wells
BaseCallerContext bc;
bc.mask = &mask;
bc.SetKeyAndFlowOrder(opts, wells.FlowOrder(), wells.NumFlows());
bc.chip_subset.InitializeChipSubsetFromOptArgs(opts, mask.W(), mask.H());
// Sampling options may reset command line arguments & change context
bc_params.InitializeSamplingFromOptArgs(opts, bc.chip_subset.NumWells());
bc_params.SetBaseCallerContextVars(bc);
ClassifyAndSampleWells(bc, bc_params.GetSamplingOpts());
// *** Setup for different datasets
BarcodeDatasets datasets_calibration(bc.run_id, bc_params.GetFiles().calibration_panel_file);
datasets_calibration.SetIonControl(bc.run_id);
datasets_calibration.GenerateFilenames("IonControl","basecaller_bam",".basecaller.bam",bc_params.GetFiles().output_directory);
BarcodeDatasets datasets(bc.run_id, bc_params.GetFiles().lib_datasets_file);
// Check if any of the template barcodes is equal to a control barcode
if (datasets_calibration.DatasetInUse())
datasets.RemoveControlBarcodes(datasets_calibration.json());
datasets.GenerateFilenames("Library","basecaller_bam",".basecaller.bam",bc_params.GetFiles().output_directory);
BarcodeDatasets datasets_tf(bc.run_id);
datasets_tf.SetTF(bc.process_tfs);
datasets_tf.GenerateFilenames("TF","basecaller_bam",".basecaller.bam",bc_params.GetFiles().output_directory);
BarcodeDatasets datasets_unfiltered_untrimmed(datasets);
BarcodeDatasets datasets_unfiltered_trimmed(datasets);
// *** Initialize remaining modules of BaseCallerContext
vector<string> bam_comments;
BaseCallerFilters filters(opts, bam_comments, bc.run_id, bc.flow_order, bc.keys, mask);
bc.filters = &filters;
BaseCallerMetricSaver metric_saver(opts, bc.chip_subset.GetChipSizeX(), bc.chip_subset.GetChipSizeY(), bc.flow_order.num_flows(),
bc.chip_subset.GetRegionSizeX(), bc.chip_subset.GetRegionSizeY(), bc_params.GetFiles().output_directory);
bc.metric_saver = &metric_saver;
// Calibration modules
bc.recalibration.Initialize(opts, bc.flow_order);
bc.recalModel.Initialize(opts, bam_comments, bc.run_id, bc.chip_subset);
// initialize the per base quality score generator - dependent on calibration
bc.quality_generator.Init(opts, chip_type, bc_params.GetFiles().input_directory, bc_params.GetFiles().output_directory, bc.recalibration.is_enabled());
// Phase estimator
bc.estimator.InitializeFromOptArgs(opts, bc.chip_subset, bc.keynormalizer);
// Barcode classification
BarcodeClassifier barcodes(opts, datasets, bc.flow_order, bc.keys, bc_params.GetFiles().output_directory,
bc.chip_subset.GetChipSizeX(), bc.chip_subset.GetChipSizeY());
bc.barcodes = &barcodes;
// Make sure calibration barcodes are initialized with default parameters
BarcodeClassifier calibration_barcodes(null_opts, datasets_calibration, bc.flow_order, bc.keys,
bc_params.GetFiles().output_directory, bc.chip_subset.GetChipSizeX(), bc.chip_subset.GetChipSizeY());
bc.calibration_barcodes = &calibration_barcodes;
// Command line parsing officially over. Detect unknown options.
opts.CheckNoLeftovers();
// Save some run info into our handy json file
bc_params.SaveParamsToJson(basecaller_json, bc, chip_type);
SaveBaseCallerProgress(0, bc_params.GetFiles().output_directory);
MemUsage("RawWellsBasecalling");
//
// Step 2. Filter training and phase estimation
//
// Find distribution of clonal reads for use in read filtering:
filters.TrainClonalFilter(bc_params.GetFiles().output_directory, wells, mask, bc.polyclonal_filter);
MemUsage("ClonalPopulation");
ReportState(analysis_start_time,"Polyclonal Filter Training Complete");
// Library phasing parameter estimation
MemUsage("BeforePhaseEstimation");
if (not bc.estimator.HaveEstimates()) {
wells.OpenForIncrementalRead();
bc.estimator.DoPhaseEstimation(&wells, &mask, bc.flow_order, bc.keys, (bc_params.NumThreads() == 1));
wells.Close();
}
bc.estimator.ExportResultsToJson(basecaller_json["Phasing"]);
bc.estimator.ExportTrainSubsetToJson(basecaller_json["TrainSubset"]);
SaveJson(basecaller_json, bc_params.GetFiles().filename_json);
SaveBaseCallerProgress(10, bc_params.GetFiles().output_directory); // Phase estimation assumed to be 10% of the work
// Initialize Barcode Classifier(s) - dependent on phase estimates
bc.barcodes->BuildPredictedSignals(bc.estimator.GetAverageCF(), bc.estimator.GetAverageIE(), bc.estimator.GetAverageDR());
bc.calibration_barcodes->BuildPredictedSignals(bc.estimator.GetAverageCF(), bc.estimator.GetAverageIE(), bc.estimator.GetAverageDR());
MemUsage("AfterPhaseEstimation");
ReportState(analysis_start_time,"Phase Parameter Estimation Complete");
MemUsage("BeforeBasecalling");
//
// Step 3. Open wells and output BAM files & initialize writers
//
// Library data set writer - always
bc.lib_writer.Open(bc_params.GetFiles().output_directory, datasets, 0, bc.chip_subset.NumRegions(),
bc.flow_order, bc.keys[0].bases(), filters.GetLibBeadAdapters(),
bc_params.NumBamWriterThreads(), basecaller_json, bam_comments);
// Calibration reads data set writer - if applicable
if (bc.have_calibration_panel)
bc.calib_writer.Open(bc_params.GetFiles().output_directory, datasets_calibration, 0, bc.chip_subset.NumRegions(),
bc.flow_order, bc.keys[0].bases(), filters.GetLibBeadAdapters(),
bc_params.NumBamWriterThreads(), basecaller_json, bam_comments);
// Test fragments data set writer - if applicable
if (bc.process_tfs)
bc.tf_writer.Open(bc_params.GetFiles().output_directory, datasets_tf, 1, bc.chip_subset.NumRegions(),
bc.flow_order, bc.keys[1].bases(), filters.GetTFBeadAdapters(),
bc_params.NumBamWriterThreads(), basecaller_json, bam_comments);
// Unfiltered / unfiltered untrimmed data set writers - if applicable
if (!bc.unfiltered_set.empty()) {
bc.unfiltered_writer.Open(bc_params.GetFiles().unfiltered_untrimmed_directory, datasets_unfiltered_untrimmed, -1,
bc.chip_subset.NumRegions(), bc.flow_order, bc.keys[0].bases(), filters.GetLibBeadAdapters(),
bc_params.NumBamWriterThreads(), basecaller_json, bam_comments);
bc.unfiltered_trimmed_writer.Open(bc_params.GetFiles().unfiltered_trimmed_directory, datasets_unfiltered_trimmed, -1,
bc.chip_subset.NumRegions(), bc.flow_order, bc.keys[0].bases(), filters.GetLibBeadAdapters(),
bc_params.NumBamWriterThreads(), basecaller_json, bam_comments);
}
//
// Step 4. Execute threaded basecalling
//
time_t basecall_start_time;
time(&basecall_start_time);
pthread_mutex_init(&bc.mutex, NULL);
pthread_t worker_id[bc_params.NumThreads()];
for (int worker = 0; worker < bc_params.NumThreads(); worker++)
if (pthread_create(&worker_id[worker], NULL, BasecallerWorker, &bc)) {
printf("*Error* - problem starting thread\n");
exit (EXIT_FAILURE);
}
for (int worker = 0; worker < bc_params.NumThreads(); worker++)
pthread_join(worker_id[worker], NULL);
pthread_mutex_destroy(&bc.mutex);
time_t basecall_end_time;
time(&basecall_end_time);
//
// Step 5. Close files and print out some statistics
//
printf("\n\nBASECALLING: called %d of %u wells in %1.0lf seconds with %d threads\n\n",
filters.NumWellsCalled(), bc.chip_subset.NumWells(),
difftime(basecall_end_time,basecall_start_time), bc_params.NumThreads());
bc.lib_writer.Close(datasets, "Library");
if (bc.have_calibration_panel)
bc.calib_writer.Close(datasets_calibration, "IonControl");
if (bc.process_tfs)
bc.tf_writer.Close(datasets_tf, "Test Fragments");
filters.TransferFilteringResultsToMask(mask);
if (!bc.unfiltered_set.empty()) {
// Must happen after filters transferred to mask
bc.WriteUnfilteredFilterStatus(bc_params.GetFiles());
bc.unfiltered_writer.Close(datasets_unfiltered_untrimmed);
bc.unfiltered_trimmed_writer.Close(datasets_unfiltered_trimmed);
datasets_unfiltered_untrimmed.SaveJson(bc_params.GetFiles().unfiltered_untrimmed_directory+"/datasets_basecaller.json");
datasets_unfiltered_trimmed.SaveJson(bc_params.GetFiles().unfiltered_trimmed_directory+"/datasets_basecaller.json");
}
metric_saver.Close();
barcodes.Close(datasets);
calibration_barcodes.Close(datasets_calibration);
if (bc.have_calibration_panel) {
datasets.json()["IonControl"]["datasets"] = datasets_calibration.json()["datasets"];
datasets.json()["IonControl"]["read_groups"] = datasets_calibration.read_groups();
}
datasets.SaveJson(bc_params.GetFiles().output_directory+"/datasets_basecaller.json");
if (bc.process_tfs)
datasets_tf.SaveJson(bc_params.GetFiles().output_directory+"/datasets_tf.json");
// Generate BaseCaller.json
bc.lib_writer.SaveFilteringStats(basecaller_json, "lib", true);
if (bc.have_calibration_panel)
bc.calib_writer.SaveFilteringStats(basecaller_json, "control", false);
if (bc.process_tfs)
bc.tf_writer.SaveFilteringStats(basecaller_json, "tf", false);
time_t analysis_end_time;
time(&analysis_end_time);
basecaller_json["BaseCaller"]["end_time"] = get_time_iso_string(analysis_end_time);
basecaller_json["BaseCaller"]["total_duration"] = (int)difftime(analysis_end_time,analysis_start_time);
basecaller_json["BaseCaller"]["basecalling_duration"] = (int)difftime(basecall_end_time,basecall_start_time);
basecaller_json["Filtering"]["qv_histogram"] = Json::arrayValue;
for (int qv = 0; qv < 50; ++qv)
basecaller_json["Filtering"]["qv_histogram"][qv] = (Json::UInt64)bc.lib_writer.qv_histogram()[qv];
SaveJson(basecaller_json, bc_params.GetFiles().filename_json);
SaveBaseCallerProgress(100, bc_params.GetFiles().output_directory);
mask.WriteRaw (bc_params.GetFiles().filename_filter_mask.c_str());
mask.validateMask();
MemUsage("AfterBasecalling");
ReportState(analysis_start_time,"Basecalling Complete");
return EXIT_SUCCESS;
}
int main (int argc, const char *argv[])
{
if (argc == 1) {
printf ("BaseCallerLite - Bare bone basecaller\n");
printf ("\n");
printf ("Usage:\n");
printf ("BaseCallerLite [options]\n");
printf ("\tOptions:\n");
printf ("\t\tComing soon\n");
printf ("\n");
return 1;
}
string libKey = "TCAG";
string inputDirectory = ".";
string outputDirectory = ".";
bool singleCoreCafie = false;
BaseCallerLite basecaller;
basecaller.regionXSize = 50;
basecaller.regionYSize = 50;
basecaller.runId = "BCLTE";
basecaller.CF = 0.0;
basecaller.IE = 0.0;
basecaller.numWellsCalled = 0;
basecaller.nextRegionX = 0;
basecaller.nextRegionY = 0;
OptArgs opts;
opts.ParseCmdLine(argc, argv);
opts.GetOption(basecaller.CF, "0.0", '-', "cf");
opts.GetOption(basecaller.IE, "0.0", '-', "ie");
opts.GetOption(inputDirectory, ".", '-', "input-dir");
opts.GetOption(outputDirectory, ".", '-', "output-dir");
opts.GetOption(singleCoreCafie, "false", '-', "singlecorecafie");
int numWorkers = 2*numCores();
if (singleCoreCafie)
numWorkers = 1;
Mask mask (1, 1);
if (mask.SetMask ((inputDirectory + "/bfmask.bin").c_str()))
exit (EXIT_FAILURE);
RawWells wells (inputDirectory.c_str(),"1.wells");
//SetWellsToLiveBeadsOnly(wells,&mask);
wells.OpenForIncrementalRead();
basecaller.maskPtr = &mask;
basecaller.wellsPtr = &wells;
basecaller.rows = mask.H();
basecaller.cols = mask.W();
basecaller.flowOrder.SetFlowOrder(wells.FlowOrder(), wells.NumFlows());
basecaller.numFlows = wells.NumFlows();
basecaller.numRegionsX = (basecaller.cols + basecaller.regionXSize - 1) / basecaller.regionXSize;
basecaller.numRegionsY = (basecaller.rows + basecaller.regionYSize - 1) / basecaller.regionYSize;
basecaller.numRegions = basecaller.numRegionsX * basecaller.numRegionsY;
basecaller.libKeyFlows.assign(basecaller.numFlows,0);
basecaller.libNumKeyFlows = basecaller.flowOrder.BasesToFlows(libKey, &basecaller.libKeyFlows[0], basecaller.numFlows);
basecaller.libSFF.Open(outputDirectory+"/rawlib.sff", basecaller.numRegions,
basecaller.flowOrder, libKey);
time_t startBasecall;
time(&startBasecall);
pthread_mutex_init(&basecaller.wellsAccessMutex, NULL);
pthread_t worker_id[numWorkers];
for (int iWorker = 0; iWorker < numWorkers; iWorker++)
if (pthread_create(&worker_id[iWorker], NULL, BasecallerWorkerWrapper, &basecaller)) {
printf("*Error* - problem starting thread\n");
return 1;
}
for (int iWorker = 0; iWorker < numWorkers; iWorker++)
pthread_join(worker_id[iWorker], NULL);
pthread_mutex_destroy(&basecaller.wellsAccessMutex);
time_t endBasecall;
time(&endBasecall);
basecaller.libSFF.Close();
printf("\nBASECALLING: called %d of %d wells in %1.1f seconds with %d threads\n",
basecaller.numWellsCalled, basecaller.rows*basecaller.cols, difftime(endBasecall,startBasecall), numWorkers);
printf("Generated library SFF with %d reads\n", basecaller.libSFF.num_reads());
return 0;
}
void BaseCallerLite::BasecallerWorker()
{
while (true) {
deque<int> wellX;
deque<int> wellY;
deque<vector<float> > wellMeasurements;
pthread_mutex_lock(&wellsAccessMutex);
if (nextRegionY >= numRegionsY) {
pthread_mutex_unlock(&wellsAccessMutex);
return;
}
int currentRegionX = nextRegionX;
int currentRegionY = nextRegionY;
int currentRegion = currentRegionX + numRegionsX * currentRegionY;
int beginY = currentRegionY * regionYSize;
int beginX = currentRegionX * regionXSize;
int endY = min((currentRegionY+1) * regionYSize,rows);
int endX = min((currentRegionX+1) * regionXSize,cols);
wellsPtr->SetChunk(beginY, endY-beginY, beginX, endX-beginX, 0, numFlows);
wellsPtr->ReadWells();
for (int y = beginY; y < endY; y++) {
for (int x = beginX; x < endX; x++) {
if (!maskPtr->Match(x, y, MaskLib))
continue;
wellX.push_back(x);
wellY.push_back(y);
wellMeasurements.push_back(vector<float>());
wellMeasurements.back().resize(numFlows);
const WellData *w = wellsPtr->ReadXY(x, y);
copy(w->flowValues, w->flowValues + numFlows, wellMeasurements.back().begin());
}
}
if (currentRegionX == 0)
printf("% 5d/% 5d: ", currentRegionY*regionYSize, rows);
if (wellX.size() == 0)
printf(" ");
else if (wellX.size() < 750)
printf(". ");
else if (wellX.size() < 1500)
printf("o ");
else if (wellX.size() < 2250)
printf("# ");
else
printf("$ ");
nextRegionX++;
if (nextRegionX == numRegionsX) {
nextRegionX = 0;
nextRegionY++;
printf("\n");
}
fflush(NULL);
pthread_mutex_unlock(&wellsAccessMutex);
BasecallerRead currentRead;
DPTreephaser dpTreephaser(flowOrder);
dpTreephaser.SetModelParameters(CF, IE, 0);
// Process the data
deque<SFFEntry> libReads;
deque<int>::iterator x = wellX.begin();
deque<int>::iterator y = wellY.begin();
deque<std::vector<float> >::iterator measurements = wellMeasurements.begin();
for (; x != wellX.end() ; x++, y++, measurements++) {
if (!maskPtr->Match(*x, *y, (MaskType)(MaskLib|MaskKeypass), MATCH_ALL))
continue;
libReads.push_back(SFFEntry());
SFFEntry& readResults = libReads.back();
stringstream wellNameStream;
wellNameStream << runId << ":" << (*y) << ":" << (*x);
readResults.name = wellNameStream.str();
readResults.clip_qual_left = 4; // TODO
readResults.clip_qual_right = 0;
readResults.clip_adapter_left = 0;
readResults.clip_adapter_right = 0;
readResults.flowgram.resize(numFlows);
int minReadLength = 8; // TODO
currentRead.SetDataAndKeyNormalize(&(measurements->at(0)), numFlows, &libKeyFlows[0], libNumKeyFlows - 1);
dpTreephaser.NormalizeAndSolve5(currentRead, numFlows); // sliding window adaptive normalization
readResults.n_bases = 0;
for (int iFlow = 0; iFlow < numFlows; iFlow++) {
readResults.flowgram[iFlow] = 100 * currentRead.solution[iFlow];
readResults.n_bases += currentRead.solution[iFlow];
}
if(readResults.n_bases < minReadLength) {
libReads.pop_back();
continue;
}
bool isFailKeypass = false;
for (int iFlow = 0; iFlow < (libNumKeyFlows-1); iFlow++)
if (libKeyFlows[iFlow] != currentRead.solution[iFlow])
isFailKeypass = true;
if(isFailKeypass) {
libReads.pop_back();
continue;
}
readResults.flow_index.reserve(readResults.n_bases);
readResults.bases.reserve(readResults.n_bases);
readResults.quality.reserve(readResults.n_bases);
unsigned int prev_used_flow = 0;
for (int iFlow = 0; iFlow < numFlows; iFlow++) {
for (hpLen_t hp = 0; hp < currentRead.solution[iFlow]; hp++) {
readResults.flow_index.push_back(1 + iFlow - prev_used_flow);
readResults.bases.push_back(flowOrder[iFlow]);
readResults.quality.push_back(20); // BaseCallerLite is stripped of QV generator
prev_used_flow = iFlow + 1;
}
}
}
libSFF.WriteRegion(currentRegion,libReads);
}
}
void DoThreadedSignalProcessing ( CommandLineOpts &inception_state, ComplexMask &from_beadfind_mask, char *chipType,
ImageSpecClass &my_image_spec, SlicedPrequel &my_prequel_setup,SeqListClass &my_keys, bool pass_tau,
BkgFitterTracker *bkg_fitter_tracker)
{
MemUsage ( "StartingBackground" );
time_t init_start;
time ( &init_start );
bool restart = not inception_state.bkg_control.restart_from.empty();
BkgFitterTracker GlobalFitter ( my_prequel_setup.num_regions );
const std::string wellsFile = string(inception_state.sys_context.wellsFilePath) + "/" + inception_state.sys_context.wellsFileName;
MakeDecisionOnGpuMultiFlowFit(inception_state);
if( restart ){
GlobalFitter = *bkg_fitter_tracker;
}
else {
GlobalFitter.global_defaults.flow_global.SetFlowOrder ( inception_state.flow_context.flowOrder ); // @TODO: 2nd duplicated code instance
// Build everything
SetBkgModelGlobalDefaults ( GlobalFitter.global_defaults, inception_state.bkg_control,chipType,inception_state.sys_context.GetResultsFolder() );
// >does not open wells file<
fprintf(stdout, "Opening wells file %s ... ", wellsFile.c_str());
RawWells preWells ( inception_state.sys_context.wellsFilePath, inception_state.sys_context.wellsFileName );
fprintf(stdout, "done\n");
CreateWellsFileForWriting ( preWells,from_beadfind_mask.my_mask, inception_state, NUMFB,
inception_state.flow_context.GetNumFlows(), my_image_spec.rows, my_image_spec.cols, chipType );
// build trace tracking
GlobalFitter.SetUpTraceTracking ( my_prequel_setup, inception_state, my_image_spec, from_beadfind_mask );
GlobalFitter.AllocateRegionData(my_prequel_setup.region_list.size());
}
TinyInitializeUglyStaticForSignalProcessing ( GlobalFitter.global_defaults , inception_state);
// plan (this happens whether we're from-disk or not):
GlobalFitter.PlanComputation ( inception_state.bkg_control );
// do we have a wells file?
ION_ASSERT( isFile(wellsFile.c_str()), "Wells file "+ wellsFile + " does not exist" );
RawWells rawWells ( inception_state.sys_context.wellsFilePath, inception_state.sys_context.wellsFileName );
// plan (this happens whether we're from-disk or not):
GlobalFitter.ThreadedInitialization ( rawWells, inception_state, from_beadfind_mask, inception_state.sys_context.GetResultsFolder(), my_image_spec,
my_prequel_setup.smooth_t0_est,my_prequel_setup.region_list, my_prequel_setup.region_timing, my_keys, restart);
MemUsage ( "AfterBgInitialization" );
time_t init_end;
time ( &init_end );
fprintf ( stdout, "InitModel: %0.3lf sec.\n", difftime ( init_end,init_start ) );
// Image Loading thread setup to grab flows in the background
// ImageTracker constructed to load flows
// must contact the GlobalFitter data that it will be associated with
// from thin air each time:
ImageTracker my_img_set ( inception_state.flow_context.getFlowSpan(),inception_state.img_control.ignoreChecksumErrors,inception_state.img_control.doSdat,inception_state.img_control.total_timeout );
my_img_set.SetUpImageLoaderInfo ( inception_state, from_beadfind_mask, my_image_spec );
my_img_set.DecideOnRawDatsToBufferForThisFlowBlock();
my_img_set.FireUpThreads();
// Now do threaded solving, going through all the flows
GlobalFitter.SpinUp();
// need to have initialized the regions for this
GlobalFitter.SetRegionProcessOrder ();
// determine maximum beads in a region for gpu memory allocations
GlobalFitter.DetermineMaxLiveBeadsAndFramesAcrossAllRegionsForGpu();
// ideally these are part of the rawWells object itself
int write_well_flow_interval = inception_state.bkg_control.saveWellsFrequency*NUMFB; // goes with rawWells
int flow_to_write_wells = -1000; // never happens unless we set it to happen
// process all flows...
// using actual flow values
Timer flow_block_timer;
Timer signal_proc_timer;
for ( int flow = inception_state.flow_context.startingFlow; flow < (int)inception_state.flow_context.endingFlow; flow++ )
{
if ((flow % NUMFB) == 0)
flow_block_timer.restart();
// coordinate with the ImageLoader threads for this flow to be read in
// WaitForFlowToLoad guarantees all flows up this one have been read in
my_img_set.WaitForFlowToLoad ( flow );
// ----- handle set up for processing this flow before we do anything needing
bool last_flow = ( ( flow ) == ( inception_state.flow_context.GetNumFlows()- 1 ) ); // actually the literal >last< flow, not just the flow in a chunk, so we can handle not having a full chunk.
// always write intervals starting at wherever we are starting
// logic here: open wells file at startingFlow, tell at what flow we need to write things out.
if (NeedToOpenWellChunk(flow-inception_state.flow_context.startingFlow, write_well_flow_interval))
{
// chunk size is flow interval unless we run out of things to do in this interval
int chunk_depth = FigureChunkDepth(flow,inception_state.flow_context.endingFlow,write_well_flow_interval);
OpenExistingWellsForOneChunk(rawWells,flow,chunk_depth); // start
flow_to_write_wells = flow+chunk_depth-1;
}
// done with set up for anything this flow needs
signal_proc_timer.restart();
// computation that modifies data
GlobalFitter.ExecuteFitForFlow ( flow,my_img_set,last_flow ); // isolate this object so it can carry out actions in any order it chooses.
ApplyClonalFilter ( *from_beadfind_mask.my_mask, inception_state.sys_context.GetResultsFolder(), GlobalFitter.sliced_chip,inception_state.bkg_control.enableBkgModelClonalFilter, flow );
// no more computation
signal_proc_timer.elapsed();
fprintf ( stdout, "SigProc: pure compute time for flow %d: %.1f sec.\n", flow, signal_proc_timer.elapsed());
MemUsage ( "Memory_Flow: " + ToStr ( flow ) );
// capture the regional parameters every 20 flows, plus one bead per region at "random"
// @TODO replace with clean hdf5 interface for sampling beads and region parameters
GlobalFitter.DumpBkgModelRegionInfo ( inception_state.sys_context.GetResultsFolder(),flow,last_flow );
GlobalFitter.DumpBkgModelBeadInfo ( inception_state.sys_context.GetResultsFolder(),flow,last_flow, inception_state.bkg_control.debug_bead_only>0 );
WriteSampleRegion(inception_state.sys_context.GetResultsFolder(), GlobalFitter, flow, inception_state.bkg_control.region_vfrc_debug);
// variables should be >captured< at the end of fitting
// and then the hdf5 dump happens across all threads as we synchronize
GlobalFitter.all_params_hdf.IncrementalWrite ( flow, last_flow );
// done capturing parameters, close out this flow
// logic here: wells file knows when it needs to write something out
if (flow==flow_to_write_wells)
WriteOneChunkAndClose(rawWells);
// Needed for 318 chips. Decide how many DATs to read ahead for every block of NUMFB flows
// also report timing for block of 20 flows from reading dat to writing 1.wells for this block
if ((flow % NUMFB) == (NUMFB - 1))
my_img_set.DecideOnRawDatsToBufferForThisFlowBlock();
// report timing for block of 20 flows from reading dat to writing 1.wells for this block
if (((flow % NUMFB) == (NUMFB - 1)) || last_flow)
fprintf ( stdout, "Flow Block compute time for flow %d to %d: %.1f sec.\n",
((flow + 1) - NUMFB), flow, flow_block_timer.elapsed());
// coordinate with the ImageLoader threads that this flow is done with
// and release resources associated with this image
// my_img_set knows what buffer is associated with the absolute flow
my_img_set.FinishFlow ( flow );
// stop GPU thread computing doing fitting of first block of flows
if (flow == (NUMFB - 1))
GlobalFitter.UnSpinMultiFlowFitGpuThreads();
}
if ( not inception_state.bkg_control.restart_next.empty() ){
string filePath = inception_state.sys_context.analysisLocation + inception_state.bkg_control.restart_next;
ofstream outStream(filePath.c_str(), ios_base::trunc);
assert(outStream.good());
//boost::archive::text_oarchive outArchive(outStream);
boost::archive::binary_oarchive outArchive(outStream);
// get region associated objects on disk first
time_t begin_save_time;
time ( &begin_save_time );
ComplexMask *from_beadfind_mask_ptr = &from_beadfind_mask;
BkgFitterTracker *GlobalFitter_ptr = &GlobalFitter;
string svn_rev = IonVersion::GetSvnRev();
outArchive
<< svn_rev
<< my_prequel_setup
<< from_beadfind_mask_ptr
<< GlobalFitter_ptr;
outStream.close();
time_t finish_save_time;
time ( &finish_save_time );
fprintf ( stdout, "Writing restart state to archive %s took %0.1f secs",
filePath.c_str(), difftime ( finish_save_time, begin_save_time ));
}
rawWells.Close();
GlobalFitter.UnSpinSingleFlowFitGpuThreads ();
TinyDestroyUglyStaticForSignalProcessing();
if ( inception_state.bkg_control.updateMaskAfterBkgModel )
from_beadfind_mask.pinnedInFlow->UpdateMaskWithPinned ( from_beadfind_mask.my_mask ); //update maskPtr
from_beadfind_mask.pinnedInFlow->DumpSummaryPinsPerFlow ( inception_state.sys_context.GetResultsFolder() );
}
