Protein secondary structure prediction using a small training set (compact model) combined with a Complex-valued neural network approach

Abstract

BackgroundProtein secondary structure prediction (SSP) has been an area of intense research interest. Despite advances in recent methods conducted on large datasets, the estimated upper limit accuracy is yet to be reached. Since the predictions of SSP methods are applied as input to higher-level structure prediction pipelines, even small errors may have large perturbations in final models. Previous works relied on cross validation as an estimate of classifier accuracy. However, training on large numbers of protein chains compromises the classifier ability to generalize to new sequences. This prompts a novel approach to training and an investigation into the possible structural factors that lead to poor predictions.Here, a small group of 55 proteins termed the compact model is selected from the CB513 dataset using a heuristics-based approach. In a prior work, all sequences were represented as probability matrices of residues adopting each of Helix, Sheet and Coil states, based on energy calculations using the C-Alpha, C-Beta, Side-chain (CABS) algorithm. The functional relationship between the conformational energies computed with CABS force-field and residue states is approximated using a classifier termed the Fully Complex-valued Relaxation Network (FCRN). The FCRN is trained with the compact model proteins.ResultsThe performance of the compact model is compared with traditional cross-validated accuracies and blind-tested on a dataset of G Switch proteins, obtaining accuracies of ∼81 %. The model demonstrates better results when compared to several techniques in the literature. A comparative case study of the worst performing chain identifies hydrogen bond contacts that lead to Coil ⇔ Sheet misclassifications. Overall, mispredicted Coil residues have a higher propensity to participate in backbone hydrogen bonding than correctly predicted Coils.ConclusionsThe implications of these findings are: (i) the choice of training proteins is important in preserving the generalization of a classifier to predict new sequences accurately and (ii) SSP techniques sensitive in distinguishing between backbone hydrogen bonding and side-chain or water-mediated hydrogen bonding might be needed in the reduction of Coil ⇔ Sheet misclassifications.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-016-1209-0) contains supplementary material, which is available to authorized users.

HSCVFNT: Inference of Time-Delayed Gene Regulatory Network Based on Complex-Valued Flexible Neural Tree Model

“…Because of its automatic features extraction, cross layer connections and flexible activation functions, the FNT model performed better than many classical neural networks and has been applied widely for solving forecasting and classification problems [ 53 , 54 , 55 , 56 , 57 ]. Due to fact that a complex-valued neural network is more flexible and functional, a complex-valued flexible neural tree (CVFNT) model, as the extension of a real-valued FNT model, is proposed to infer the time-delayed regulations in GRN [ 58 , 59 , 60 ]. A tree-structural based encoding method with a specific instruction set is selected for representing a CVFNT model.…”

Multifaceted analysis of training and testing convolutional neural networks for protein secondary structure prediction

“…For example, we observed that if our model is trained and tested on data sets with 50% sequence identity instead of 25%, it has 1.3 points higher accuracy (top of Fig 4). This issue has been raised several times by different groups [34,46,90,91]. In this work, we developed a rigorous protocol that relies on two passes of two different sequence alignment software programs (HHblits and Clustal-Omega) to enforce 25% identity within our data sets.…”

“…One common issue that has recurred repeatedly during the history of secondary structure prediction is that reported accuracies have not always been upheld when methods were applied to new benchmark test sets [30,34,38,46,82,90]. Rost and Sander [34] state that overoptimistic claims are caused by inadequate quality and size of test sets not meeting several requirements.…”

“…Rost and Sander [34] state that overoptimistic claims are caused by inadequate quality and size of test sets not meeting several requirements. First, significant pairwise sequence identity cannot be observed between proteins used for training and test set [90,91]. For example, in [84] training was performed on a set representing the entire sequence space of the PDB leading to very similar proteins in training and testing sets.…”

Application of improved intelligent ant colony algorithm in protein folding prediction

“…In the two-dimensional Toy model, this paper transforms the protein folding prediction problem into a function optimal value problem, that is, how to obtain N-2 bond angles to minimize the energy value. 22 In this way, we abstract the abstraction into concrete, simplify the complex problem, and change the structure of the real protein without changing its characteristics.…”