Assessing the potential of bee-collected pollen sequence data to train machine learning models for geolocation of...

Area of Science:

• Palynology
• Forensic Science
• Bioinformatics

Background:

• Pollen analysis (palynology) aids in tracking object origins and movements.
• Limitations in pollen identification and reference data hinder precise geolocation.
• Existing DNA metabarcoding data often uses wind-pollinated species, limiting location specificity.

Purpose of the Study:

• To assess the efficacy of bee-collected pollen DNA metabarcoding for predicting sample origin.
• To evaluate supervised machine learning models for geolocation using pollen assemblages.
• To compare the performance of models trained on raw versus taxonomically classified DNA sequence data.

Main Methods:

• Compiled bee-collected pollen DNA sequence relative abundance data from western U.S. projects.
• Applied supervised machine learning models, including Random Forest and k-Nearest Neighbors.
• Trained models using both raw DNA sequence variants (ASVs) and taxonomically clustered data.

Main Results:

• Machine learning models accurately predicted sample origin based on pollen assemblages.
• Random Forest and k-Nearest Neighbors models demonstrated high accuracy and low error rates.
• Models trained on raw sequence data performed comparably to those trained on classified data, simplifying the process.

Conclusions:

• Bee-collected pollen is a valuable, underutilized resource for geolocation.
• Supervised machine learning effectively predicts sample origin using pollen DNA data.
• The study provides a framework for future geolocation efforts utilizing pollen data and machine learning.