Difference in HB plots of open and ligand-bound form show few important individual changes in tertiary hydrogen bonding pattern. Over 98% of long-range contacting residues are in close proximity of another contact, compared with 30% for non-contacting pairs. Average accuracy on long-range contacts for the element alignment predictor. 3): the segment–segment features improve the identification of contacting regions between secondary structure elements and the DNN is able to refine the prediction scores. The second step is the transfer of the first electron from NADPH via an electron transfer chain. We use an energy-based method (Nagata et al., 2011) to assign energies then probabilities to the alignment between contacting secondary structure elements and derive approximate probabilities of contact for their residue pairs. The receptive fields used in the simulations results are essentially square patches (Supplementary Fig. The feature vector for segment has the following components: Three vectors (20 entries each) representing the average amino acid distribution computed over the profiles of Sn – 1, Sn and Sn + 1. Average Acc and for seq. Examining the HB plot of the closed and open state of CYP2B4 revealed that the rearrangement of tertiary hydrogen bonds was in excellent agreement with the current knowledge of the cytochrome P450 catalytic cycle. The four alignment predictors are also trained using 10-fold cross-validation on the data described in Section 2.2. Here, we introduce several new ideas for contact prediction using primarily a multi-stage machine learning approach, with increasingly refined levels of resolution. The integration over time provided by the different levels in the stack corresponds to the intuition that folding is a somewhat organized, non-instantaneous, process which proceeds through successive stages of refinement. Thus, as in the CASP experiments, we focus primarily on long-range contact prediction. However, the accuracy of current contact predictors often barely exceeds 20% on long-range contacts, falling short of the level required for ab initio structure prediction.. The effectiveness of this approach results from the fact that a mutation in position i of a protein is more likely to be associated with a mutation in position j than with a back-mutation in i if both positions are functionally coupled (e.g. In our experiments, we obtain considerable better overall performance by increasing this percentage to 20% (data not shown). Furthermore, over 60% of contacting pairs are in the proximity of at least 10 different contacts, compared with 2.5% for non-contacting pairs (Supplementary Fig. Conversely, the sparse long-range contacts are the most informative and also the most difficult to predict. Then, the probability represents the probability of contact for the pair (i and j; Supplementary Fig. There are three types of purely spatial input features: residue–residue features coarse features and alignment features. The only method (Wang et al., 2010) outperforming CMAPpro on the CASP9 dataset by a small margin relies on 3D structure models for deriving contact predictions through consensus, which defeats the purpose of predicting contact maps from scratch. Recently, a new elegant mutual information-based measure for correlated mutations, PSICOV, has been proposed in Jones et al. and sep. on CASP9 set. The input of the 2D-BRNN for the pair Sn and Sm consists of two feature vectors and as well as the number of elements between and . Each has two different kinds of input features: purely spatial features and temporal features. For strand–strand contacts, the phase values alternate between 0 and 1, whereas for helix–helix contacts, the phase values cycle periodically from 0 to 6 (Supplementary Fig. These networks are then trained by on-line backpropagation for one epoch. The lengths (two entries) in residues of the intervals between Sn – 1 and Sn and Sn and Sn + 1. Thus, in practice, at each training epoch, we append a new neural network to the growing DNN, initialize it with the weights of the previous level and train it by back-propagation using the true contacts as the targets (or softer targets could be derived from folding data). The performance on the strand–strand regions, E–E, E–E and helix–helix regions, H–H, H–H, are obtained by using the contact probabilities in Equations (10), (12), (11) and (13), respectively. Globally, the two predictors assign a high probability of contact (grey dots) to approximately the same regions. For long-range contacts, the accuracy of the new CMAPpro predictor is close to 30%, a significant increase over existing approaches. Here, we have introduced a new approach for the prediction of protein contact maps. The blue and red dots represent the correctly and incorrectly predicted contacts, respectively, among the L top-scored residue pairs. Indeed, if we remove the only three TBM domains from the CASP9 dataset and focus exclusively on the FM targets, which are harder to predict, then RR490’s accuracy (L/5) drops down from 0.32 to 0.28, whereas CMAPpro’s accuracy increases from 0.31 to 0.32. Author Summary Protein contact prediction and contact-assisted folding has made good progress due to direct evolutionary coupling analysis (DCA). A protein contact map represents the distance between all possible amino acid residue pairs of a three-dimensional protein structure using a binary two-dimensional matrix. In this way, training and validation sets share neither sequence nor structural similarities. The protein domains having <5 contacts in the strand–strand and helix–helix regions have been excluded from the evaluation. Accuracy (L/5 long range contacts) versus network depth for the set of test domains (All), the test domains of length between 50 and 100 residues (50–100, 87 domains), between 101 and 150 (100–150 and 111 domains), between 151 and 200 (150–200 and 76 domains) and longer than 200 (200, 90 domains). Parallel contact (P), Anti-parallel contact (A) and No-contact (N) are the three classes considered by the coarse contact and orientation predictor.