DeepFun is a deep-learning-based model for functional evaluation of genetic variants at the single-base resolution. Specifically, DeepFun is a comprehensive collection of chromatin profiles from ENCODE and Roadmap projects, which constructs the feature space, including 1548 DNase I accessibility, 1536 histone mark, and 4795 transcription factor binding profiles covering 225 tissues or cell lines. With such comprehensive epi-genomics annotations and the pre-trained model, this web-server provides an online service to quickly assess the impact of genetic variants in a tissue- or cell-type-specific manner.
For a list of query variants across all chromatin features, the screen analysis computes SNP Activity (accessibility or binding probability) Difference (SAD) or relative log fold change of odds (log-odds) difference between two alleles.
In silico saturated mutagenesis analysis
Perform an in silico saturated mutagenesis analysis of the regions surrounding 200 bp of one query variant under a target chromatin feature of interest.
Ruifeng Hu, Guangsheng Pei, Peilin Jia, Zhongming Zhao @ CPH, UTHealth-Houston SBMI.
DeepFun is free and open to all users and there is no login requirement.
Pei, G., Hu, R., Dai, Y., Manuel, A.M., Zhao, Z. and Jia, P. (2021) Predicting regulatory variants using a dense epigenomic mapped CNN model elucidated the molecular basis of trait-tissue associations. Nucleic Acids Res, 49, 53-66.