eDNAssay: A machine learning tool that accurately predicts qPCR cross-amplification

Area of Science:

• Ecology
• Genetics
• Bioinformatics

Background:

• Environmental DNA (eDNA) sampling offers a cost-effective wildlife monitoring method using quantitative PCR (qPCR) assays.
• Ensuring qPCR assay specificity is challenging when closely related species coexist, often requiring extensive in vitro testing.
• Current in silico methods for predicting assay specificity lack accuracy due to poorly understood mismatch requirements.

Purpose of the Study:

• To improve the accuracy of in silico assessments for qPCR assay specificity.
• To streamline the development of specific qPCR assays for environmental DNA monitoring.
• To reduce the bottleneck of in vitro testing in assay development.

Main Methods:

• Trained random forest classifiers using 530 specificity tests from 10 qPCR assays and 82 synthetic gene fragments.
• Utilized SYBR Green intercalating dye (n=262) and TaqMan hydrolysis probes (n=268) for specificity testing.
• Validated model performance on six independent assays not included in training.

Main Results:

• The primer-only model achieved 99.6% accuracy in cross-validation (SYBR Green).
• The full-assay model (TaqMan probes) demonstrated 100% accuracy in cross-validation.
• Independent validation showed 92.4% accuracy for the primer-only model and 96.5% for the full-assay model.

Conclusions:

• Accurate in silico prediction of qPCR assay specificity is achievable using machine learning.
• These models enable faster and more confident development of specific eDNA assays.
• An online tool, eDNAssay, is available for predicting qPCR cross-amplification.