[paper] 3Cnet

3Cnet: Pathogenicity prediction of human variants using knowledge transfer with deep recurrent neural networks

... many rare diseases are caused by the missense mutations

... PP3 is one of the standards in the ACMG guideline which means in-silico assessment criterion.

... REVEL developed a random forest algorithm that incorporates various pathogenicity predictors to build an integrated predictor for missense variants.

... CADD is another ensemble method which uses linear regression to integrate different scoring tools.

... FATHMM makes use of evolutionary information to recognize evolutionarily well-conserved variants.

... VEST4, POLYPHEN2, and SIFT are other well known prediction tools to predict the functionality change based on Random Forest, Naive Bayes, and statistical mehods, respectively.

... Circularity cas lead to an overlap between training and evaluation datasets and consequently result in overfitting

... deep neural networks require massive amounts of data for effective training compared to random forests.

... Clinical data consist of pathogenic variants and benign variants reported to ClinVar based on the ACMG guideline.

... the pathogenicity classifier is composed of two fully connected layers. The first FC layer is expected to extract the difference between wild-type and mutant as a feature vector. a dropout layer was applied to the first layer to avoid overfitting.