Performance and Limitations of Out-Of-Distribution Detection for Insect DNA Barcoding.

Area of Science:

• Genomics
• Bioinformatics
• Ecology

Background:

• DNA barcoding is vital for species identification in biological surveys.
• Underexplored regions often have incomplete reference databases, leading to potential misidentification of unknown species.
• Detecting these 'out-of-distribution' samples is crucial but their reliability is not well-studied.

Purpose of the Study:

• To evaluate the effectiveness of DNA barcoding for detecting out-of-distribution samples.
• To assess the impact of sequence noise and fragment length on detection performance.
• To provide guidelines for robust DNA barcoding identification procedures.

Main Methods:

• Utilized extensively sampled insect datasets with sufficient intra-species genetic variation.
• Assessed performance of out-of-distribution detection using genetic distance and deep learning metrics.
• Tested identification and detection accuracy with varying sequence quality and fragment lengths.

Main Results:

• DNA barcoding identification remains accurate even with short, noisy fragments.
• Out-of-distribution detection performance degrades with sequence noise and lack of diagnostic characters.
• Significant performance reduction in out-of-distribution detection was observed for fragments <300 bp, regardless of method.

Conclusions:

• While DNA barcoding is robust for identification, out-of-distribution detection requires careful consideration of fragment length and sequence quality.
• Short DNA fragments (<300 bp) severely compromise the ability to detect unknown species.
• Guidelines are provided to enhance the reliability of DNA barcoding in biodiversity assessments, especially in data-deficient regions.