Improved model quality assessment using deep neural networks

Protein structure modeling is crucial for a detailed understanding of the biological function at the molecular level (Wallner&Elofsson, 2003; Wang et al, 2009). In 2003 we developed the first single-model quality estimation program ProQ (Wallner&Elofsson, 2003). In contrast to earlier methods ProQ is not trained to recognise the native structure but to estimate the quality of a model. In ProQ2 profile weights were added to improve the predictions (Ray et al, 2012; Uziela&Wallner, 2016) and in ProQ3 (Uziela et al, 2016) we added energy terms calculated from Rosetta (Leaver-Fay, 2011). The ProQ methods have since their introduction been the best single-model based quality assessors in CASP (Kryshtafovych et al, 2016).

In the last few years machine learning using so called deep neural networks has proven to be clearly superior to other machine learning methods. We find that using identical inputs as in ProQ2 and ProQ3 but replacing the support vector machine with a deep neural network a substantial improvement can be obtained for both ProQ2 and ProQ3.


Wallner B, Elofsson A (2003) “Can correct protein models be identified?” Protein Sci 12(5), 1073-1086
Wang Z, Tegge AN, Cheng J (2009) “Evaluating the absolute quality of a single protein model using structural features and support vector machines.” Proteins 75(3), 638-647
Ray A, Lindahl E, Wallner B (2012) “Improved model quality assessment using ProQ2.” BMC Bioinformatics 13, 224
Uziela K, Wallner B (2016) “ProQ2: estimation of model accuracy implemented in Rosetta.” Bioinformatics 32(9), 1411-1413
Uziela K, Shu N, Wallner B, Elofsson A (2016) “ProQ3: Improved model quality assessments using Rosetta energy terms.” Sci Rep 6, 33509
Leaver-Fay A et al (2011) “ROSETTA3: an object-oriented software suite for the simulation and design of macromolecules.” Methods Enzymol 487, 545-574
Kryshtafovych A et al (2016) “Methods of model accuracy estimation can help selecting the best models from decoy sets: Assessment of model accuracy estimations in CASP11.” Proteins 84 Suppl 1, 349-369
Uziela, K., Menendez Hurtado, D., Shu, N., Wallner, B. and Elofsson, A. (Epub 2017) ProQ3D: improved model quality assessments using deep learning. Bioinformatics (Epub ahead of print)