The paper introduces SpliceAI, a deep neural network that accurately predicts splice junctions from pre-mRNA sequences, enabling precise prediction of noncoding genetic variants causing cryptic splicing. The network, trained on GENCODE-annotated pre-mRNA transcripts, achieves 95% top-k accuracy in predicting splice junctions. It can also predict the impact of genetic variants on splicing, with a validation rate of 75% in RNA-seq data. The study finds that cryptic splice variants, often overlooked in clinical practice, are a significant cause of rare genetic disorders, particularly in neurodevelopmental disorders. These variants are enriched in patients with autism and intellectual disability, and their effects are strongly deleterious in the human population. The research highlights the importance of considering noncoding regions in the diagnosis and treatment of genetic diseases.The paper introduces SpliceAI, a deep neural network that accurately predicts splice junctions from pre-mRNA sequences, enabling precise prediction of noncoding genetic variants causing cryptic splicing. The network, trained on GENCODE-annotated pre-mRNA transcripts, achieves 95% top-k accuracy in predicting splice junctions. It can also predict the impact of genetic variants on splicing, with a validation rate of 75% in RNA-seq data. The study finds that cryptic splice variants, often overlooked in clinical practice, are a significant cause of rare genetic disorders, particularly in neurodevelopmental disorders. These variants are enriched in patients with autism and intellectual disability, and their effects are strongly deleterious in the human population. The research highlights the importance of considering noncoding regions in the diagnosis and treatment of genetic diseases.