Temporal Dynamics of Genetic Line for Forensic Genealogy Using Time Series Prediction

Jianwei Jin^*

• Kunming University of Science and Technology, Kunming, China

Recent advances in forensic investigative genetic genealogy (FIGG) have highlighted the critical need to model temporal dynamics in genetic lineages for improved population structure resolution, ancestry inference, and the identification of unknown individuals. However, most current methods rely on static genomic representations, overlooking the dynamic nature of genetic evolution and its effects on trait heritability and population drift. Traditional approaches, such as PCA and fixed-cluster admixture models, often fail to capture fine-scale temporal shifts, limiting their effectiveness in longitudinal forensic analysis. To overcome these limitations, we introduce a forecasting framework that integrates predictive modeling with biologically grounded genetic representations. Our system consists of EvoCast, a dual-stream temporal neural architecture, and GenScope, a tailored forecasting strategy. EvoCast encodes convolutional genomic features and time-series phenotypic trajectories via cross-attention, enabling interpretable latent trajectory modeling while preserving genetic structure. GenScope enhances forecasting with mutation-consistent latent drift, heritability-aware trait adjustments, and graph-based genotypic imputation, supporting the simulation of evolutionary dynamics across generations. Empirical results on both synthetic and real datasets show that EvoCast with GenScope surpasses baseline models in predictive accuracy and interpretability, particularly under population structure shifts, missing genotype data, and long-range inference scenarios. By aligning computational modeling with biological priors and evolutionary principles, our framework provides a scalable and robust solution to advance the precision and reliability of FIGG applications.