Deep convolutional networks for quality assessment of protein folds.

摘要: Motivation The computational prediction of a protein structure from its sequence generally relies on method to assess the quality models. Most assessment methods rank candidate models using heavily engineered structural features, defined as complex functions atomic coordinates. However, very few have attempted learn these features directly data. Results We show that deep convolutional networks can be used predict ranking model structures solely basis their raw three-dimensional densities, without any feature tuning. develop neural network performs par with state-of-the-art algorithms literature. is trained decoys CASP7 CASP10 datasets and performance tested CASP11 dataset. Additional testing CASP12, CAMEO 3DRobot confirms consistently well across variety structures. While learns globally does not rely predefined it analyzed implicitly identifies regions deviate native structure. Availability implementation code are available at https://github.com/lamoureux-lab/3DCNN_MQA. Supplementary information data Bioinformatics online.