predicted contacts and 3D models of 510 non-redundant membrane proteins with solved structures in PDBTM

1) The contacts are predicted by the deep learning method developed by Dr. Jinbo Xu. Each protein has a .gcnn file, which is a text file containing a L*L matrix. Meanwhile, L is the protein sequence length, each entry in the matrix is a predicted probability of the corresponding residue pair forming a contact. Please ignore those entries (i,j) where abs(i-j)<6 . 2) The 3D models generated by CNS software from predicted contacts and secondary structure. Each predicted 3D model is a PDB file. For each protein, 5 models are predicted.