DANN is a deep learning approach for annotating the pathogenicity of genetic variants. It improves upon the Combined Annotation-Dependent Depletion (CADD) method, which uses a linear kernel support vector machine (SVM) to annotate variants. However, SVMs cannot capture non-linear relationships among features, which can limit performance. DANN uses a deep neural network (DNN) to better capture these relationships, achieving a 19% relative reduction in error rate and a 14% increase in area under the curve (AUC) compared to CADD's SVM. DANN uses a feature set and training data similar to CADD, but employs deep learning techniques such as dropout and momentum training to accelerate DNN training. It also uses Compute Unified Device Architecture (CUDA)-compatible graphics processing units (GPUs) for faster training. DANN outperforms simpler linear approaches like logistic regression (LR) and SVMs for classification problems with many features and samples. The model uses 949 features for each variant, including real-valued numbers, integers, and binary values. Training data consists of 16,627,775 observed variants and 49,407,057 simulated variants. The data is split into training, validation, and testing sets in an 8:1:1 ratio. DANN achieves a classification accuracy of 66.1%, outperforming SVM (58.2%) and LR (59.8%). DANN's performance is attributed to its ability to capture non-linear relationships among features, which are important in genome annotations. However, its accuracy is still unsatisfactory, possibly due to mislabeled training data, insufficient features for functional prediction, and the need for further model training. DANN also performs well in distinguishing pathogenic ClinVar variants from likely benign ESP alleles. It is the most useful annotation algorithm for non-coding variants, which make up the majority of human genetic variation. DANN is expected to play an important role in prioritizing putative causal variants for further analysis. The study was supported by the National Institute of Biomedical Imaging and Bioengineering. No conflicts of interest were declared.DANN is a deep learning approach for annotating the pathogenicity of genetic variants. It improves upon the Combined Annotation-Dependent Depletion (CADD) method, which uses a linear kernel support vector machine (SVM) to annotate variants. However, SVMs cannot capture non-linear relationships among features, which can limit performance. DANN uses a deep neural network (DNN) to better capture these relationships, achieving a 19% relative reduction in error rate and a 14% increase in area under the curve (AUC) compared to CADD's SVM. DANN uses a feature set and training data similar to CADD, but employs deep learning techniques such as dropout and momentum training to accelerate DNN training. It also uses Compute Unified Device Architecture (CUDA)-compatible graphics processing units (GPUs) for faster training. DANN outperforms simpler linear approaches like logistic regression (LR) and SVMs for classification problems with many features and samples. The model uses 949 features for each variant, including real-valued numbers, integers, and binary values. Training data consists of 16,627,775 observed variants and 49,407,057 simulated variants. The data is split into training, validation, and testing sets in an 8:1:1 ratio. DANN achieves a classification accuracy of 66.1%, outperforming SVM (58.2%) and LR (59.8%). DANN's performance is attributed to its ability to capture non-linear relationships among features, which are important in genome annotations. However, its accuracy is still unsatisfactory, possibly due to mislabeled training data, insufficient features for functional prediction, and the need for further model training. DANN also performs well in distinguishing pathogenic ClinVar variants from likely benign ESP alleles. It is the most useful annotation algorithm for non-coding variants, which make up the majority of human genetic variation. DANN is expected to play an important role in prioritizing putative causal variants for further analysis. The study was supported by the National Institute of Biomedical Imaging and Bioengineering. No conflicts of interest were declared.