Experiences of social isolation among older adults with type 2 diabetes mellitus: A qualitative study.

De novo mutations (DNMs) are DNA sequence alterations that arise spontaneously in germ cells or during early embryonic development []. DNMs play a crucial pathogenic role in a broad spectrum of genetic and complex diseases, particularly in early-onset neurological disorders such as autism spectrum disorder [], intellectual disability [], early-onset Parkinson's disease [], and early-onset Alzheimer’s disease [], more precisely, coding-region DNMs contributed approximately 30% of the diagnosis rate of individual cases and explained pathogenic mutations in 45% of female patients []. In autism, DNMs can account for 52% to 67% of low-risk families and 30% to 39% of the pathogenic sources of all cases []. In severe developmental disorders (DD), Pathogenic DNMs are significantly enriched in development-related genes, resulting in a birth prevalence ranging from 1/213 to 1/448 []. The widespread adoption of high-throughput sequencing technologies [] has greatly expanded the discovery of DNMs; however, the functional and pathogenic significance of most variants remains elusive. Therefore, accurately assessing the pathogenic potential of DNMs is essential for molecular diagnosis, etiological interpretation, and personalized medicine. In recent years, the rapid advancement of machine learning and deep learning has led to the development of numerous variant pathogenicity prediction tools. SpliceAI [] employs deep neural networks to capture genomic splicing signals; CADD [] integrates conservation metrics, regulatory annotations, and transcript information. More recent studies have further advanced the field toward more expressive and biologically informed models. For example, SVPath [] demonstrates the importance of variant-type-specific modeling for exon structural variants; V2P [] introduces phenotype-conditioned multi-task learning for joint prediction of pathogenicity and disease outcomes; GNN-based approaches such as GNN-MAP [] leverage graph representations to integrate multimodal annotations; and popEVE [] combines deep evolutionary signals with human population variation to improve calibration of variant effect prediction. These developments highlight a shift toward phenotype-aware, multimodal, and better calibrated pathogenicity prediction frameworks.