The paper introduces DeepSEA, a deep learning-based algorithm for predicting the effects of noncoding variants on chromatin. Noncoding variants, which account for a majority of disease-associated SNPs, are challenging to characterize functionally. DeepSEA directly learns a regulatory sequence code from large-scale chromatin-profiling data, enabling single-nucleotide sensitivity in predicting chromatin effects. The model integrates sequence information from a wide context, learns sequence code at multiple spatial scales, and uses a multitask architecture to share predictive strength across different chromatin factors. Evaluations show that DeepSEA accurately predicts chromatin features and variant effects, outperforming existing methods. It also effectively prioritizes functional variants, including eQTLs and disease-associated SNPs, without relying on additional annotations. The method's ability to predict insertions and deletions (indels) further enhances its utility. DeepSEA is a powerful tool for uncovering the regulatory information in noncoding regions and understanding the functions of complex traits and diseases.The paper introduces DeepSEA, a deep learning-based algorithm for predicting the effects of noncoding variants on chromatin. Noncoding variants, which account for a majority of disease-associated SNPs, are challenging to characterize functionally. DeepSEA directly learns a regulatory sequence code from large-scale chromatin-profiling data, enabling single-nucleotide sensitivity in predicting chromatin effects. The model integrates sequence information from a wide context, learns sequence code at multiple spatial scales, and uses a multitask architecture to share predictive strength across different chromatin factors. Evaluations show that DeepSEA accurately predicts chromatin features and variant effects, outperforming existing methods. It also effectively prioritizes functional variants, including eQTLs and disease-associated SNPs, without relying on additional annotations. The method's ability to predict insertions and deletions (indels) further enhances its utility. DeepSEA is a powerful tool for uncovering the regulatory information in noncoding regions and understanding the functions of complex traits and diseases.