Ultra Model Sets 11 - 14
DOWNLOAD === https://bltlly.com/2tdfef
Previously we developed an accurate contact prediction method and a novel feature encoding technique to maximize contact prediction accuracy on a novel multi-task learning architecture. Both challenges have now been successfully resolved. Firstly, we have improved our contact prediction model on the DeepCovid convolutional recurrent neural networks (CNNs) architecture and integrated with our contact-assisted template guided model (CATSEM) to model long-range interactions using both non-template based distance restraints and contact-assisted folding. Secondly, our feature encoding technique is based on the novel steerable pyramid representation of signed distance function (SP-SDF), which can better represent long-range geometric and distance information. It directly maps long-range contacts into feature vectors and, thus, the feature extraction layer of our approach is not affected by the specific lattice and resolution. This enables us to achieve state-of-the-art accuracy on our benchmarks and also creates a highly versatile template-assisted contact prediction framework, e.g., it can be easily used on many other kinds of lattices and/or resolutions. The entire code is released as an open source package to the community as well.
In the previous work, we trained a novel method for estimating the structural similarity score for pairwise residue contacts. This is useful to predict contacts between sequences with no similarity. However, as we discussed in our previous work, this method is not yet accurate enough as contact prediction on our benchmarks. Therefore, we further improved this method by constructing a deep neural network called DMNN (Deep Model for Multiple Neural Network) to predict contacts between sequences with a relatively low percentage of sequence identity. All contact predictions can be downloaded from the web server freely for further test and applications. DMNN can also be used to predict homo-oligomeric contacts when sequence similarity is low. d2c66b5586