Protein Contact Map Prediction
RaptorX Contact Predict is a web server that predicts contact map of a protein sequence without using any template information. It applies group graphical lasso (GGL) to contact prediction that integrates joint multi-family evolutionary coupling (EC) analysis and supervised machine learning and works particularly well on proteins without many sequence homologs. Our joint EC analysis is unique in that it not only uses residue coevolution information in the target protein family, but also the residue coevolution information in the related families, which may have divergent sequences but similar folds. Our supervised learning predicts contacts by making use of sequence and sequence profile, contact (distance) potential and other information. Our experimental results show that this server can obtain ~0.92 L/10 accuracy for long-range contacts, and ~0.85 L/10 accuracy for medium-range contacts on 150 protein families used by PSICOV. Finally, the tertiary structure of the input sequence is generated by the aid of predicted contacts, but without using any templates.