Copy number variants (CNVs) are pivotal in driving phenotypic variation that facilitates species adaptation. They are significant contributors to various disorders, making ancient genomes cru- cial for uncovering the genetic origins of disease susceptibility across populations. However, detecting CNVs in ancient samples poses substantial challenges due to several factors. Ancient DNA (aDNA) is often highly degraded, and this degradation is further complicated by contamination from micro- bial DNA and DNA from closely related species, introducing additional noise into sequencing data. Finally, the typically low coverage of aDNA renders accurate CNV detection particularly difficult. Conventional CNV calling algorithms, optimized for high coverage and long reads, often underper- form in such conditions. To address these limitations, we introduce LYCEUM, a deep learning-based CNV caller specifically designed for low-coverage aDNA. LYCEUM performs transfer learning from a model designed to detect CNVs in another noisy data domain, whole exome sequencing then it per- forms fine-tuning with a few aDNA samples for which semi-ground truth CNV calls are available. Our findings demonstrate that LYCEUM accurately identifies CNVs even in highly downsampled genomes, maintaining robust performance across a range of coverage levels. Thus, LYCEUM offers researchers a reliable solution for CNV detection in challenging ancient genomic datasets.

This repository contains processed samples, groundtruth data, and CNV predictions for real and simulated datasets to reproduce the analyses in the paper: LYCEUM: Learning to Call Copy Number Variants on Low-Coverage Ancient Genomes.