Abstract: The genetic function of most non-coding regions is unclear, and many genetic variants have been found in these regions. How to identify associated disease variants is still a challenge. A Support Vector Machine based algorithm CADD has been proposed, which can annotate coding and non-coding region variants. However, CADD fails to capture non-linear relationship among features. To solve this problem, this paper designed a hybrid convolutional neural network and fully connected neural network model. This model can capture non-linear relationship well among features. Our method achieves the highest accuracy of 66.44% on the testing set.

Key words: deep learning, genetic variants, pathogenicity, annotation