Merge branch 'DOR-696_store_qs_as_float' into 'master'

DOR-696 Change qs tag type to float Closes DOR-696 See merge request machine-learning/dorado!1005
nanoporetech · May 17, 2024 · d6b0f68 · d6b0f68
2 parents c54efac + 661ddd7
commit d6b0f68
Show file tree

Hide file tree

Showing 4 changed files with 7 additions and 7 deletions.
diff --git a/documentation/SAM.md b/documentation/SAM.md
@@ -25,7 +25,7 @@
 |        |                                                            |
 | ------ | -----------------------------------------------------------|
 | RG:Z:  | `<runid>_<basecalling_model>_<barcode_arrangement>`        |
-| qs:i:  | mean basecall qscore rounded to the nearest integer        |
+| qs:f:  | mean basecall qscore                                       |
 | ts:i:  | the number of samples trimmed from the start of the signal |
 | ns:i:  | the basecalled sequence corresponds to the interval `signal[ts : ns]` <br /> the move table maps to the same interval. <br /> note that `ns` reflects trimming (if any) from the rear <br /> of the signal. |
 | mx:i:  | read mux                                                   |

diff --git a/dorado/read_pipeline/messages.cpp b/dorado/read_pipeline/messages.cpp
@@ -37,8 +37,8 @@ std::string ReadCommon::generate_read_group() const {
 }
 
 void ReadCommon::generate_read_tags(bam1_t *aln, bool emit_moves, bool is_duplex_parent) const {
-    int qs = static_cast<int>(std::round(calculate_mean_qscore()));
-    bam_aux_append(aln, "qs", 'i', sizeof(qs), (uint8_t *)&qs);
+    float qs = calculate_mean_qscore();
+    bam_aux_append(aln, "qs", 'f', sizeof(qs), (uint8_t *)&qs);
 
     float du = (float)(get_raw_data_samples() + num_trimmed_samples) / (float)sample_rate;
     bam_aux_append(aln, "du", 'f', sizeof(du), (uint8_t *)&du);
@@ -108,8 +108,8 @@ void ReadCommon::generate_read_tags(bam1_t *aln, bool emit_moves, bool is_duplex
 }
 
 void ReadCommon::generate_duplex_read_tags(bam1_t *aln) const {
-    int qs = static_cast<int>(std::round(calculate_mean_qscore()));
-    bam_aux_append(aln, "qs", 'i', sizeof(qs), (uint8_t *)&qs);
+    float qs = calculate_mean_qscore();
+    bam_aux_append(aln, "qs", 'f', sizeof(qs), (uint8_t *)&qs);
     uint32_t duplex = 1;
     bam_aux_append(aln, "dx", 'i', sizeof(duplex), (uint8_t *)&duplex);
 

diff --git a/dorado/summary/summary.cpp b/dorado/summary/summary.cpp
@@ -168,7 +168,7 @@ bool SummaryData::write_rows_from_reader(
         auto duration = reader.get_tag<float>("du");
 
         auto seqlen = reader.record->core.l_qseq;
-        auto mean_qscore = reader.get_tag<int>("qs");
+        auto mean_qscore = reader.get_tag<float>("qs");
 
         auto num_samples = reader.get_tag<int>("ns");
         auto trim_samples = reader.get_tag<int>("ts");

diff --git a/tests/ReadTest.cpp b/tests/ReadTest.cpp
@@ -36,7 +36,7 @@ TEST_CASE(TEST_GROUP ": Test tag generation", TEST_GROUP) {
         REQUIRE(alignments.size() == 1);
         bam1_t* aln = alignments[0].get();
 
-        CHECK(bam_aux2i(bam_aux_get(aln, "qs")) == 14);
+        CHECK(bam_aux2f(bam_aux_get(aln, "qs")) == 14.0f);
         CHECK(bam_aux2i(bam_aux_get(aln, "ns")) == 4132);
         CHECK(bam_aux2i(bam_aux_get(aln, "ts")) == 132);
         CHECK(bam_aux2i(bam_aux_get(aln, "mx")) == 2);