Multivariate statistical approach and machine learning for the evaluation of biogeographical ancestry inference in the forensic field

Abstract

The biogeographical ancestry (BGA) of a trace or a person/skeleton refers to the component of ethnicity, constituted of biological and cultural elements, that is biologically determined. Nowadays, many individuals are interested in exploring their genealogy, and the capability to distinguish biogeographic information about population groups and subgroups via DNA analysis plays an essential role in several fields such as in forensics. In fact, for investigative and intelligence purposes, it is beneficial to inference the biogeographical origins of perpetrators of crimes or victims of unsolved cold cases when no reference profile from perpetrators or database hits for comparative purposes are available. Current approaches for biogeographical ancestry estimation using SNPs data are usually based on PCA and Structure software. The present study provides an alternative method that involves multivariate data analysis and machine learning strategies to evaluate BGA discriminating power of unknown samples using different commercial panels. Starting from 1000 Genomes project, Simons Genome Diversity Project and Human Genome Diversity Project datasets involving African, American, Asian, European and Oceania individuals, and moving towards further and more geographically restricted populations, powerful multivariate techniques such as Partial Least Squares-Discriminant Analysis (PLS-DA) and machine learning techniques such as XGBoost were employed, and their discriminating power was compared. PLS-DA method provided more robust classifications than XGBoost method, showing that the adopted approach might be an interesting tool for forensic experts to infer BGA information from the DNA profile of unknown individuals, but also highlighting that the commercial forensic panels could be inadequate to discriminate populations at intra-continental level.

Exploring the complexity of African populations variability with Machine Learning

“…The idea is to identify a ML data analysis workflow, starting from data visualization up to supervised and unsupervised classification, supporting the work of anthropologists to identify sex and ancestry from human cranial remains. Recent works support our vision, highlighting the importance to introduce ML in Anthropology [20]- [24].…”

How Physical and Molecular Anthropology Interplay in the Creation of Biological Profiles of Unidentified Migrants

“…The obtained results highlight that most of the unknown individuals fall into the East Africa cluster, except for Samples 105-2, 154-1154-2, and 178-3 that fall into West Africa. However, since PCA modeling is not a reliable discrimination/classification model [ 62 ] and the individuals from Central and South Africa were missing from the database used, the ancestry inference, especially at an intra-continental level, may not be accurate. The difficulty of ancestry inference at an intra-continental level was also confirmed by the results of uniparental markers.…”

“…Despite this, as shown in Figure 4 and Figure 5 where the comparison between anthropological (physical and molecular) results relating to sex and ancestry are summarized, the molecular analysis support the physical data in both sex and ancestry inference, confirming the presence of 21 male crania of African origin. As proposed by Alladio et al [ 62 ] and further developed by Pilli et al [ 63 ], to infer intra-continental biogeographical ancestry in a forensic context, we suggest adopting a well-characterized and well-defined database and a robust classification method such as a multivariate statistical and machine learning approach associated with selected markers (AIMSNPs ancestry informative markers).…”

Forensic DNA Phenotyping: a review on SNP panels, genotyping techniques, and prediction models

“…In 2001, Jobling published the rst review that considered Y-SNP haplogroup inferring as an exclusion tool to target an initial suspect [5]. Many reviews that discuss the aspects of BGA inference such as the development of panels and selecting classi cation algorithms are available [47,48,[67][68][69][70]. However, only few discuss the different panels for FDP application [7,13,49], and they are usually centred around the most used ones.…”

“…Both methods may provide similar accuracy when the same SNP panel is used [15] although ML methods require a higher computational cost and expertise. Indeed, several articles compared and introduced different classi ers for FDP analysis [12,48,67,68,70,117,205,206,217,227,252,299,327].…”