Merge branch 'DOR-736_short_read_crashfix' into 'master'

DOR-736 Add handling for very short reads that are entirely trimmed by the ScalerNode Closes DOR-736 See merge request machine-learning/dorado!1047
nanoporetech · Jun 3, 2024 · b531918 · b531918
2 parents f9beb39 + 1d389d0
commit b531918
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 3 deletions.
diff --git a/dorado/read_pipeline/ScalerNode.cpp b/dorado/read_pipeline/ScalerNode.cpp
@@ -141,7 +141,8 @@ void ScalerNode::input_thread_fn() {
         int trim_start = 0;
         if (is_rna_model) {
             trim_start = determine_rna_adapter_pos(*read, m_model_type);
-            if (m_trim_rna_adapter) {
+            if (m_trim_rna_adapter &&
+                size_t(trim_start) < read->read_common.get_raw_data_samples()) {
                 read->read_common.raw_data =
                         read->read_common.raw_data.index({Slice(trim_start, at::indexing::None)});
                 read->read_common.rna_adapter_end_signal_pos = 0;
@@ -241,8 +242,12 @@ void ScalerNode::input_thread_fn() {
                         utils::DEFAULT_TRIM_MIN_ELEMENTS);
             }
 
-            read->read_common.raw_data =
-                    read->read_common.raw_data.index({Slice(trim_start, at::indexing::None)});
+            if (size_t(trim_start) < read->read_common.get_raw_data_samples()) {
+                read->read_common.raw_data =
+                        read->read_common.raw_data.index({Slice(trim_start, at::indexing::None)});
+            } else {
+                trim_start = 0;
+            }
         }
 
         read->read_common.num_trimmed_samples = trim_start;

diff --git a/tests/data/pod5/degenerate/trimming_bomb.pod5 b/tests/data/pod5/degenerate/trimming_bomb.pod5
diff --git a/tests/test_simple_basecaller_execution.sh b/tests/test_simple_basecaller_execution.sh
@@ -78,6 +78,9 @@ if $dorado_bin basecaller $model_5k_v43 $data_dir/duplex/pod5 --modified-bases 5
 fi
 set -e
 
+# Check that dorado handles degenerate reads without crashing
+$dorado_bin basecaller $model_5k_v43 $data_dir/pod5/degenerate --skip-model-compatibility-check > $output_dir/error_condition.fq
+
 echo dorado summary test stage
 $dorado_bin summary $output_dir/calls.bam