Machine learning approaches outperform distance- and tree-based methods for DNA barcoding of Pterocarpus wood

Area of Science:

• Botany
• Genetics
• Forensic Science

Background:

• DNA barcoding is crucial for identifying wood species, particularly for combating illegal logging and ensuring trade compliance.
• Developing accurate and efficient analytical methods is essential for the broad application of DNA barcoding in wood trade and forensic science.
• Existing distance- and tree-based methods have limitations in identification accuracy and cost-effectiveness for wood species discrimination.

Purpose of the Study:

• To compare the performance of machine-learning approaches (MLAs) against traditional distance- and tree-based methods for DNA barcoding of Pterocarpus wood.
• To evaluate the identification accuracy and cost-effectiveness of different DNA barcodes and MLA classifiers.
• To assess the utility of MLAs for the forensic identification of CITES-listed Pterocarpus species.

Main Methods:

• Collected and curated a reference dataset of 205 DNA sequences from four barcodes (ITS2, matK, ndhF-rpl32, rbcL) for six commercial Pterocarpus species.
• Applied various machine-learning algorithms (BLOG, BP-neural network, SMO, J48), distance-based (TaxonDNA), and tree-based (NJ tree) methods for species identification.
• Evaluated methods based on identification accuracy and cost-effectiveness, and tested discrimination of Pterocarpus santalinus from P. tinctorius.

Main Results:

• Machine-learning approaches demonstrated superior identification accuracy (30.8-100%) compared to distance- (15.1-97.4%) and tree-based methods (11.1-87.5%).
• The SMO classifier performed best among the machine learning algorithms.
• The combination of ITS2 + matK barcodes with the SMO classifier achieved 100% resolution for discriminating the six Pterocarpus species.
• Single barcode ndhF-rpl32 with MLAs successfully discriminated the CITES-listed Pterocarpus santalinus from P. tinctorius.

Conclusions:

• Machine-learning approaches offer higher accuracy and cost-effectiveness for DNA barcoding of Pterocarpus wood compared to traditional methods.
• MLAs are a promising tool for accurate species-level identification in wood forensics and trade.
• The ITS2 + matK barcode combination with SMO classifier is highly effective for Pterocarpus species discrimination, while ndhF-rpl32 is suitable for forensic identification of specific CITES-listed species.