Predicting haplogroups using a versatile machine learning program (PredYMaLe) on a new mutationally balanced 32 Y-STR...

Area of Science:

• Forensic Genetics
• Population Genetics
• Bioinformatics

Background:

• Short Tandem Repeat (STR) markers are crucial for human population genetics and forensic identification.
• Predicting Y-chromosome haplogroups from Y-STR profiles aids in tracing paternal lineage and understanding population history.
• Assessing the impact of STR mutability on haplogroup prediction accuracy is essential for robust forensic tools.

Purpose of the Study:

• To develop a mutationally well-balanced 32 Y-STR multiplex (CombYplex) and a machine learning (ML) program (PredYMaLe).
• To evaluate the impact of STR mutability on haplogroup prediction accuracy using CombYplex and PredYMaLe.
• To assess the performance of ML classifiers for predicting haplogroup classes from Y-STR profiles.

Main Methods:

• Designed CombYplex with two sub-panels (M1 and M2) featuring average and high-mutation STRs.
• Tested CombYplex on 996 human samples to evaluate its discrimination capacity.
• Developed and applied ML classifiers (SVM, Random Forest, Neural Network, Bayesian) using PredYMaLe to predict haplogroups from Y-STR profiles.

Main Results:

• CombYplex demonstrated sufficient resolution to separate haplogroup classes.
• Support Vector Machine (SVM) and Random Forest classifiers achieved high prediction accuracy (average 97%), outperforming other ML methods.
• Haplogroup assignation accuracy varied, with smaller sample sizes (e.g., E1b1b, G) showing lower scores, which improved with more training data.

Conclusions:

• The ML approach, combined with well-balanced Y-STR panels and large training sets, is a robust method for accurate haplogroup prediction.
• Further research into confounding factors like CNV-STR and gene conversion can further optimize prediction scores.
• CombYplex and PredYMaLe offer a promising tool for forensic genetics and population studies, enhancing Y-chromosome haplogroup analysis.