Abstracts of Invited Lectures – Moses Schanfield Memorial Symposium
Presentation number: IL 37
Abstract number: 124-ISABS-2026
ARTIFICIAL INTELLIGENCE ENABLES SCALE, CONSISTENCY, AND RIGOR IN FORENSIC IDENTITY INFERENCE
Budowle Bruce1,2
1Department of Forensic Medicine, University of Helsinki, Helsinki, Finland; 2Othram, The Woodlands, TX, United States of America
b.budowle@att.net
For more than two decades forensic science took a targeted approach of typing relatively small panels of short tandem repeat (STR) markers coupled with capillary electrophoresis for human identification purposes. However, this approach has limitations such as sensitivity of detection, particularly with highly degraded DNA samples, and resolution power only for direct comparisons and kinship analyses typically with first degree relationships. Massively parallel sequencing (MPS) and dense single nucleotide polymorphisms (SNPs) analyses greatly extends human identification capabilities. MPS coupled with forensic genetic genealogy (FGG) overcomes many of the limitations of STR typing. To establish potential kinship relationships, dense SNP data are searched against a database(s) of reference samples from consented volunteers. Associations are made primarily on identity -by-descent segment analysis, with the amount and total size of shared segments serving as indicators of genetic relationships. By FGG, searching for near and distant relatives greatly expands the range of cases in which DNA evidence can generate investigative leads. With these capabilities there is a need to go beyond predominantly human-centered workflows and limited hypothesis testing and instead embrace automation and capabilities to reason consistently , transparently, and at scale over increasingly complex genetic, genealogical, and contextual information. Artificial intelligence (AI) will be an enabling layer which is particularly suited for FGG as a computational decision-support system(s) that structures, prioritizes, and documents reasoning over genetic associations, genealogical structures, and investigative context during identity hypothesis development in a sustainably scaling manner. Incorporation of FGG and AI requires governance mechanisms that ensure transparency, accountability, privacy protection, and human oversight.
Keywords: FGG, forensic genetic genealogy, artificial intelligence, single nucleotide polymorphisms, kinship
Presentation number: IL 38
Abstract number: 122-ISABS-2026
AI AS A TIME MACHINE: INFERRING ANCIENT MOLECULAR PHENOTYPES
Capra John A. (Tony)1
1University of California, San Francisco, CA, United States of America
tony@capralab.org
The sequencing of genomes from archaic hominins and modern humans has transformed our understanding of recent human history. However, due to the difficulty of inferring phenotypes from genotypes, ancient DNA has yielded few insights into the traits of ancient individuals. In this talk, I will describe how my group is using powerful machine learning methods to infer molecular phenotypes from ancient genetic sequences. I will illustrate how leveraging sequence-based deep neural networks can quantify protein structures, gene expression, splicing, and genome three -dimensional (3D) structure in archaic individuals. These analyses have revealed substantial similarities and differences between modern and ancient individuals that highlight molecular divergence in systems relevant to known phenotypic differences in the immune, metabolic, and skeletal systems. We identify specific loci where modern Eurasians have inherited novel molecular phenotypes from Neanderthal ancestors and show that these provide putative molecular mechanisms for phenotypes associated with the introgressed haplotypes. In summary, deep learning applied to ancient DNA sequences has great potential to reveal previously unobservable molecular differences between humans and our closest relatives.
Keywords: AI, human evolution, Neanderthals, ancient DNA, gene regulation
Presentation number: IL 39
Abstract number: 118-ISABS-2026
NEANDERTALS IN FOCUS: WHAT NEW GENOMES TELL US ABOUT THEIR BIOLOGY AND INTERACTIONS WITH HUMANS
Hajdinjak Mateja1
1Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany
mateja_hajdinjak@eva.mpg.de
Despite genome-wide data being recovered from close to 20,000 ancient humans to date, genomic data from Neandertals, our closest evolutionary relatives, are still comparatively sparse. Thus far, nuclear genetic data have been recovered from 35 Neandertals from 17 archaeological sites, spanning large parts of their history and geographical range. These data have offered a broad overview of Neandertal populations, indicating the existence of multiple distinct groups, as well repeated population turnovers. Archaeological and genetic evidence suggests that Neandertals lived in small groups, however, less is known if these groups were part of isolated communities or belonged to larger, well-connected populations. Here, I will present our recent efforts in reconstructing a more fine-scale view of Neandertal populations, focusing on some of the very last Neandertals from ten archaeological sites in Belgium and France. We used minimally destructive sampling of 36 skeletal remains which were radiocarbon dated to between ~52 and ~36 kya. Out of those, 27 contained enough endogenous DNA to generate autosomal, mitochondrial and Y-chromosomal data, including a new high-coverage genome from a 45,000-year-old Neandertal from Goyet. We found that late Neandertals from west Eurasia had much higher genetic diversity than the Neandertals from the Altai mountains, which lived between ~130 and 60 kya, suggesting that they lived in larger or better-connected groups. They also had fewer long tracts of homozygosity, comparable to those found in the Upper Palaeolithic humans present in Europe around the same time. And although these Neandertals overlapped temporally with early modern humans, we find no evidence of recent gene-flow from humans in their genomes. Moreover, we find no evidence for the accumulation of the deleterious mutations in the genomes of late Neandertals and no increase in genetic load over time, suggesting that genetics played a minor role in Neandertal extinction.
Keywords: neandertals, ancient DNA, admixture
Presentation number: IL 40
Abstract number: 143-ISABS-2026
PROBABILISTIC INTELLIGENCE FOR MITOCHONDRIAL DNA MIXTURE DECONVOLUTION
Holland Mitchell1, McElhoe Jennifer1
1Biochemistry & Molecular Biology Department, Pennsylvania State University, State College, PA, United States of America
mmh20@psu.edu
MixtureAce MT™ is a user environment for analysis of PCR-targeted mtDNA sequence data. The software runs in a cloud/Windows environment and accommodates fastq or bam files from MPS systems. Noise from sequencing error and sequences arising from nuclear mtDNA segments (NUMTs) are reduced. Contributor proportions and haplogroups are generated using the probabilistic technique of Vohr et al. [FSIG 2017]. Both the contributor proportions and haplogroup of each contributor is estimated simultaneously. Files previously generated were combined from two sources to form in-silico mixtures. GeneMarker HTS generated alignment files (bam files) for each sole-source contributor were used to create mixtures, combined in various ratios (1:1, 1:3, 1:9, 1:49, and 1:99) with each mixture consisting of a total 0.5, 1, 2, 5, or 9K read fragments. Contributor proportions and haplogroups were estimated simultaneously. The process is phylogenetically directed in that individual reads are assigned likelihoods of belonging to each individual haplogroup. The software identifies the best-supported haplogroups as those receiving the most support from read likelihoods and counts. MixtureAce MT™ reports the top-most supported haplogroups and the contribution proportion of the best-supported haplogroups. The haplogroup calls are substantiated by a listing of the phylogenetically diagnostic variant positions (diagnostic SNPs, single nucleotide polymorphisms) that point to the haplogroup call and provide a partial haplotype from the haplogroup defining SNPs. The points of view are those of the authors and do not reflect the views of Penn State University, West Virginia University, or the National Institute of Justice. Any mention of commercial products is done for scientific transparency and should not be viewed as an endorsement of the product or manufacturer.
Keywords: forensic genetics, mitochondrial DNA, haplogroup classification, MPS, probabilistic modeling
Presentation number: IL 41
Abstract number: 125-ISABS-2026
AI-DRIVEN INSIGHTS INTO THE GENETIC BASIS OF HUMAN FACIAL VARIATION
Kayser Manfred1
1Erasmus MC University Medical Center Rotterdam, Rotterdam, Netherlands
m.kayser@erasmusmc.nl
The enormous – near-individual-specific – variability of human faces has long intrigued researchers across diverse disciplines and the broader public. Although facial morphology is highly heritable, elucidating its genetic determinants is challenging due to the large phenotypic and underlying genetic complexity. In this presentation, I will summarize a series of collaborative efforts aimed at characterizing the genetic basis of human facial appearance, spanning from the first genome-wide association study (GWAS) that identified five genetic loci, via our recently published GWAS reporting almost 200 loci, to ongoing efforts, including those based on AI. I will demonstrate that, beyond the expected gains from increased sample sizes and expanded phenotypic characterization, the Combined GWAS (C-GWAS) framework we developed substantially enhanced locus discovery. I will present how these genetic insights can be leveraged to predict facial appearance from genetic data. I will show how AI-based facial phenotyping and C-GWAS of AI-derived facial traits enables the identification of additional loci not detectable from conventionally collected phenotypes. I will highlight how explainable AI allows the visualization of the identified genetic effects on the facial image. Finally, I will outline ongoing efforts to further refine the genetic understanding of facial variation through large-scale C-GWAS encompassing nearly one thousand facial traits in more than 50,000 individuals.
Keywords: face, genome-wide association study, genetic face prediction, AI-derived facial traits, explainable AI
Presentation number: IL 42
Abstract number: 166-ISABS-2026
THE GENETIC HISTORY OF THE WESTERN BALKANS AND THE GENETIC IMPACT OF THE SLAVIC MIGRATIONS
Krause Johannes1
1Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany
krause@eva.mpg.de
The second half of the first millennium CE was a period of profound cultural, demographic, and political transformation across Central and Eastern Europe, commonly associated with the emergence and expansion of Slavic-speaking populations. While textual and archaeological evidence has long suggested large-scale changes during this period, the extent to which these developments were driven by migration, cultural diffusion, or Slavicisation has remained debated. Here, we synthesize recent genome-wide data from 555 ancient individuals, including 359 individuals from Slavic-associated archaeological contexts dating as early as the seventh century CE, to investigate the demographic impact of the Slavic migrations on the Western Balkans and wider Central-Eastern Europe. The results demonstrate substantial population movement originating from Eastern Europe between the sixth and eighth centuries CE, leading to the replacement of more than 80% of the local gene pool in regions such as Eastern Germany, Poland, and Croatia. Genetic evidence further reveals strong biological links among Slavic-period populations across geographically distant regions, including Croatia, Eastern Germany, and Poland–Northwestern Ukraine. These groups share extensive identical-by-descent (IBD) segments with one another while showing minimal genetic continuity with preceding local populations, supporting a recent shared origin and rapid demographic expansion. In contrast, Roman and Migration Period individuals from Croatia cluster genetically with present-day Italian and Eastern Mediterranean populations, highlighting the scale of the demographic transformation associated with the Slavic expansion. Despite this broad pattern of genetic turnover, significant regional variability is evident. The absence of strongly sex-biased admixture suggests that migration involved entire communities rather than predominantly male warrior groups, while varying levels of admixture indicate differing degrees of assimilation with local autochthonous populations. In Eastern Germany, genetic changes coincided with shifts in social organization, including intensified inter and intra-site relatedness and increased patrilocality. However, Slavic-period societies in the Northern Balkans appear to have retained aspects of earlier Migration Period social structures, suggesting that pre existing cultural practices persisted despite large-scale demographic and linguistic change. Overall, the integration of archaeological and genetic evidence supports the interpretation that the widespread cultural and linguistic transformations observed across Europe between the sixth and eighth centuries CE were closely linked to large-scale population movements associated with the Slavic migrations.
Keywords: population genetics, slavic migrations, ancient DNA, demographic transformation, central and eastern Europe
Presentation number: IL 43
Abstract number: 131-ISABS-2026
BRIDGING GENOMIC AND EPIGENOMIC DATA VIA MACHINE LEARNING FOR FORENSIC DNA PHENOTYPING
Pośpiech Ewelina1
1Department of Genomics and Forensic Genetics, Pomeranian Medical University in Szczecin, Szczecin, Poland
ewelina.pospiech@pum.edu.pl
Forensic DNA Phenotyping (FDP) serves as an intelligence tool for perpetrator profiling when there are no suspects and no matches in national forensic DNA databases. The success of FDP became remarkable with the development of prediction models for eye, hair, and skin color based on the analysis of a finite set of SNP markers. Significant progress has also been achieved in predicting height, freckles, hair shape, and alopecia. However, the prediction of certain progressive appearance traits depends on age, which is an important factor in the models being developed. Notably, age can now be effectively estimated using DNA methylation analysis. In a study of ~1,000 individuals, we showed that incorporating age significantly improves hair graying prediction compared to SNP -only approaches. Integration of genetic and epigenetic data is also essential for facial skin aging traits, which are only partly heritable (~50%), with the remaining variation likely driven by epigenetic factors. Within the EPIGENOME project, using genome-wide SNP and methylation microarray data from >700 Polish individuals, we developed predictive models for wrinkle formation and perceived age (i.e., age estimated from facial appearance). These achieve an accuracy of ±0.4% of facial surface area for wrinkles and 3.3 years for perceived age. Since DNA methylation reflects both genetic and environmental influences, it also enables behavioral profiling. A promising example is BMI prediction. While existing models often rely on hundreds of variables, we developed a compact, forensically oriented model combining SNPs and CpG methylation, achieving an accuracy of ±2.9 BMI units. In summary, combining genomic and epigenomic data provides substantial opportunities for further development of FDP. At the same time, the role of machine learning methods, capable of handling large datasets in the context of relatively small sample sizes, is crucial for marker selection and model development.
Keywords: forensic DNA phenotyping, epigenetic and genetic data, facial skin aging, hair greying, prediction modelling
Presentation number: IL 44
Abstract number: 158-ISABS-2026
PREDICTING GEOLOCATION FROM VIRAL DNA USING AI
Keskitalo Martta1, Heino Matti1, Ge Jianye2, Budowle Bruce1,2,3, Mari Toppinen1, Sajantila Antti1,4
1Department of Forensic Medicine, University of Helsinki, Helsinki, Finland; 2Othram Inc., The Woodlands, TX, United States of America; 3Forensic Science Institute, Radford University, Radford, VA, United States of America; 4Forensic Medicine Unit, Finnish Institute for Health and Welfare, Helsinki, Finland
antti.sajantila@helsinki.fi
Persistent human DNA viruses are globally distributed, typically acquired early in life, and leave stable genomic traces in host tissues, thereby illustrating viral evolution and reflecting relationships among human populations. This viral “fingerprint” may complement human genomic data as a biogeographic marker, particularly in settings where reference datasets are limited or fragmented. We assembled 24,441 near‑full‑length genomes from NCBI GenBank, representing 29 human DNA viruses across five families (Polyomaviridae, Herpesviridae, Papillomaviridae, Parvoviridae, and Hepadnaviridae) with metadata extracted from database records and primary literature. Phylogenetic analyses were conducted using BEAST2, and machine ‑learning models were implemented through the mGPS framework to infer the geographic origin of viral sequences. Phylogenetic reconstructions revealed continent‑ and country‑level clustering for numerous viruses, including BKPyV, JCPyV, HHV1, HHV2, HHV3, HHV6B, HCMV, HPV16, HPV18, HPV31, HPV45, and HBV. Geographic structure was strongest in regions with sufficient sequence representation, whereas global data availability remained highly uneven. Sixteen viruses met the sample‑size criteria for mGPS modelling, yielding weighted F1 scores from 0.7663 to 0.9980, indicating robust continent‑level predictivity despite dataset imbalance. Overall, several persistent human DNA viruses (14 of 29) show promise as biogeographic markers, offering a complementary tool for forensic identification and population studies. Integrating phylogenetics and machine learning enables inference of host geographic origin, although broader global sequence representation, improved metadata standards, and appropriate ethical frameworks will be essential for future applications.
Keywords: forensic genetics, geolocation, human DNA viruses, persistent DNA viruses, ancestry markers
Published: June 16th, 2026.
Copyright: © 2026 MSIL symposium authors. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
