Quality Of Structure Prediction Research Articles

Evaluating DCA-based method performances for RNA contact prediction by a well-curated data set.

RNA molecules play many pivotal roles in a cell that are still not fully understood. Any detailed understanding of RNA function requires knowledge of its three-dimensional structure, yet experimental RNA structure resolution remains demanding. Recent advances in sequencing provide unprecedented amounts of sequence data that can be statistically analyzed by methods such as direct coupling analysis (DCA) to determine spatial proximity or contacts of specific nucleic acid pairs, which improve the quality of structure prediction. To quantify this structure prediction improvement, we here present a well curated data set of about 70 RNA structures of high resolution and compare different nucleotide–nucleotide contact prediction methods available in the literature. We observe only minor differences between the performances of the different methods. Moreover, we discuss how robust these predictions are for different contact definitions and how strongly they depend on procedures used to curate and align the families of homologous RNA sequences.

A further leap of improvement in tertiary structure prediction in CASP13 prompts new routes for future assessments.

We present our assessment of tertiary structure predictions for hard targets in Critical Assessment of Structure Prediction round 13 (CASP13). The analysis includes (a) assignment and discussion of best models through scores-aided visual inspection of models for each evaluation unit (EU); (b) ranking of predictors resulting from this evaluation and from global scores; and (c) evaluation of progress, state of the art, and current limitations of protein structure prediction. We witness a sizable improvement in tertiary structure prediction building on the progress observed from CASP11 to CASP12, with (a) top models reaching backbone RMSD <3 å for several EUs of size <150 residues, contributed by many groups; (b) at least one model that roughly captures global topology for all EUs, probably unprecedented in this track of CASP; and (c) even quite good models for full, unsplit targets. Better structure predictions are brought about mainly by improved residue-residue contact predictions, and since this CASP also by distance predictions, achieved through state-of-the-art machine learning methods which also progressed to work with slightly shallower alignments compared to CASP12. As we reach a new realm of tertiary structure prediction quality, new directions are proposed and explored for future CASPs: (a) dropping splitting into EUs, (b) rethinking difficulty metrics probably in terms of contact and distance predictions, (c) assessing also side chains for models of high backbone accuracy, and (d) assessing residue-wise and possibly residue-residue quality estimates.

Discovery of Novel ncRNA Sequences in Multiple Genome Alignments on the Basis of Conserved and Stable Secondary Structures.

Recently, non-coding RNAs (ncRNAs) have been discovered with novel functions, and it has been appreciated that there is pervasive transcription of genomes. Moreover, many novel ncRNAs are not conserved on the primary sequence level. Therefore, de novo computational ncRNA detection that is accurate and efficient is desirable. The purpose of this study is to develop a ncRNA detection method based on conservation of structure in more than two genomes. A new method called Multifind, using Multilign, was developed. Multilign predicts the common secondary structure for multiple input sequences. Multifind then uses measures of structure conservation to estimate the probability that the input sequences are a conserved ncRNA using a classification support vector machine. Multilign is based on Dynalign, which folds and aligns two sequences simultaneously using a scoring scheme that does not include sequence identity; its structure prediction quality is therefore not affected by input sequence diversity. Additionally, ensemble defect was introduced to Multifind as an additional discriminating feature that quantifies the compactness of the folding space for a sequence. Benchmarks showed Multifind performs better than RNAz and LocARNATE+RNAz, a method that uses RNAz on structure alignments generated by LocARNATE, on testing sequences extracted from the Rfam database. For de novo ncRNA discovery in three genomes, Multifind and LocARNATE+RNAz had an advantage over RNAz in low similarity regions of genome alignments. Additionally, Multifind and LocARNATE+RNAz found different subsets of known ncRNA sequences, suggesting the two approaches are complementary.

Automated Procedure for Contact-Map-Based Protein Structure Reconstruction

Knowledge of the three-dimensional structures of ion channels allows for modeling their conductivity characteristics using biophysical models and can lead to discovering their cellular functionality. Recent studies show that quality of structure predictions can be significantly improved using protein contact site information. Therefore, a number of procedures for protein structure prediction based on their contact-map have been proposed. Their comparison is difficult due to different methodologies used for validation. In this work, a Contact Map-to-Structure pipeline (C2S_pipeline) for contact-based protein structure reconstruction is designed and validated. The C2S_pipeline can be used to reconstruct monomeric and multimeric proteins. The median RMSD of structures obtained during validation on a representative set of protein structures, equaled 5.27 Å, and the best structure was reconstructed with RMSD of 1.59 Å. The validation is followed by a detailed case study on the KcsA ion channel. Models of KcsA are reconstructed based on different portions of contact site information. Structural feature analysis of acquired KcsA models is supported by a thorough analysis of electrostatic potential distributions inside the channels. The study shows that electrostatic parameters are correlated with structural quality of models. Therefore, they can be used to discriminate between high and low quality structures. We show that 30 % of contact information is needed to obtain accurate structures of KcsA, if contacts are selected randomly. This number increases to 70 % in case of erroneous maps in which the remaining contacts or non-contacts are changed to the opposite. Furthermore, the study reveals that local reconstruction accuracy is correlated with the number of contacts in which amino acid are involved. This results in higher reconstruction accuracy in the structure core than peripheral regions.

Funneling and frustration in the energy landscapes of some designed and simplified proteins.

We explore the similarities and differences between the energy landscapes of proteins that have been selected by nature and those of some proteins designed by humans. Natural proteins have evolved to function as well as fold, and this is a source of energetic frustration. The sequence of Top7, on the other hand, was designed with architecture alone in mind using only native state stability as the optimization criterion. Its topology had not previously been observed in nature. Experimental studies show that the folding kinetics of Top7 is more complex than the kinetics of folding of otherwise comparable naturally occurring proteins. In this paper, we use structure prediction tools, frustration analysis, and free energy profiles to illustrate the folding landscapes of Top7 and two other proteins designed by Takada. We use both perfectly funneled (structure-based) and predictive (transferable) models to gain insight into the role of topological versus energetic frustration in these systems and show how they differ from those found for natural proteins. We also study how robust the folding of these designs would be to the simplification of the sequences using fewer amino acid types. Simplification using a five amino acid type code results in comparable quality of structure prediction to the full sequence in some cases, while the two-letter simplification scheme dramatically reduces the quality of structure prediction.

Quantifying variances in comparative RNA secondary structure prediction

BackgroundWith the advancement of next-generation sequencing and transcriptomics technologies, regulatory effects involving RNA, in particular RNA structural changes are being detected. These results often rely on RNA secondary structure predictions. However, current approaches to RNA secondary structure modelling produce predictions with a high variance in predictive accuracy, and we have little quantifiable knowledge about the reasons for these variances.ResultsIn this paper we explore a number of factors which can contribute to poor RNA secondary structure prediction quality. We establish a quantified relationship between alignment quality and loss of accuracy. Furthermore, we define two new measures to quantify uncertainty in alignment-based structure predictions. One of the measures improves on the “reliability score” reported by PPfold, and considers alignment uncertainty as well as base-pair probabilities. The other measure considers the information entropy for SCFGs over a space of input alignments.ConclusionsOur predictive accuracy improves on the PPfold reliability score. We can successfully characterize many of the underlying reasons for and variances in poor prediction. However, there is still variability unaccounted for, which we therefore suggest comes from the RNA secondary structure predictive model itself.

31 Discovery of novel ncRNA by scanning multiple genome alignments

Recently, non-coding RNAs (ncRNAs) have been discovered with novel functions, and it has been appreciated that there is a pervasive transcription. Therefore, de novo computational ncRNA detection that is accurate and efficient is desirable. The purpose of this study is to develop a ncRNA detection method based on structural conservation. A new method called Multifind, based on Multilign (Xu & Mathews, 2011), was developed. It uses an algorithm that predicts common structures among multiple sequences and estimates the probability that input sequences are ncRNA using a classification support vector machine (SVM). Multilign uses Dynalign (Mathews & Turner, 2002), which folds and aligns two sequences simultaneously without requiring any sequence identity; its structure prediction quality will therefore not be affected by input sequence diversity. Benchmarks showed, Multifind performs better than RNAz on testing sequences extracted from Rfam database (Gardner et al., 2011), especially on sequences that are more diverse. For de novo ncRNA discovery in genomes, Multifind had an advantage in low similarity regions of genome alignments. Multifind takes about 48 hours to finish scanning the whole yeast genome alignment and RNAz takes about 4 hours, therefore, its computational requirements do not present a barrier for most of the users.

PPfold 3.0: fast RNA secondary structure prediction using phylogeny and auxiliary data

PPfold is a multi-threaded implementation of the Pfold algorithm for RNA secondary structure prediction. Here we present a new version of PPfold, which extends the evolutionary analysis with a flexible probabilistic model for incorporating auxiliary data, such as data from structure probing experiments. Our tests show that the accuracy of single-sequence secondary structure prediction using experimental data in PPfold 3.0 is comparable to RNAstructure. Furthermore, alignment structure prediction quality is improved even further by the addition of experimental data. PPfold 3.0 therefore has the potential of producing more accurate predictions than it was previously possible. PPfold 3.0 is available as a platform-independent Java application and can be downloaded from http://birc.au.dk/software/ppfold.

Protein secondary structure content in solution, films and tissues: Redundancy and complementarity of the information content in circular dichroism, transmission and ATR FTIR spectra

The paper presents a simple and robust method to determine protein secondary structure from circular dichroism, transmission and attenuated total reflection (ATR) Fourier transform infrared spectra. It is found that the different spectroscopic methods bring valuable but roughly identical information on the secondary structure of proteins. ATR and transmission FTIR spectra display distinct differences, yet the secondary structure can be predicted from their spectra with roughly the same success. It is also found that one wavenumber or wavelength includes the large majority of the information correlated with secondary structure content and no more than 3 significant independent wavenumbers/wavelengths could be found for any of the spectroscopic data. This finding indicates that more complex linear combinations of the absorbance or ellipticities will not further improve secondary structure predictions. Furthermore, the information content in CD, transmission and ATR FTIR spectra is largely redundant. If combining CD and FTIR results in some improvement of structure prediction quality, the improvement is too modest to prompt spectroscopists to collect different spectroscopic data for structure prediction purposes. On the other hand, the data collected show that the quality of the FTIR spectrometers is such that biosensors or imaging methods sampling from 10 − 9 to 10 − 15 g yield spectra of sufficient quality to analyze protein secondary structure. These new techniques open the way to a new area of research, both in protein conformational response to ligand and imaging at sub-cellular scales.

Fourier transform coupled to tryptophan-scanning mutagenesis: Lessons from its application to the prediction of secondary structure in the acetylcholine receptor lipid-exposed transmembrane domains

Although Fourier transform (FT) and tryptophan-scanning mutagenesis (TrpScanM) have been extremely useful for predicting secondary structures of membrane proteins, they are deemed to be low-resolution techniques. Herein, we describe the combined use of FT and TrpScanM (FT-TrpScanM) as a more reliable approach for the prediction of secondary structure. Five TrpScanM studies of the acetylcholine receptor lipid-exposed transmembrane domains (LETMDs) were revisited and analyzed by FT-TrpScanM. FT analysis of the raw data from the aforementioned TrpScanM studies supports and validates the conclusions derived from their tryptophan-periodicity profiles. Furthermore, by FT-TrpScanM, we were able to determine the minimum number of consecutive tryptophan substitutions necessary for more robust prediction of α-helical secondary structures and evaluate the quality of structure predictions by α-helical character curves. Finally, this study encourages future utilization of FT-TrpScanM to more reliably predict secondary structures of the membrane protein LETMDs.

An Information Measure of the Quality of Protein Secondary Structure Prediction

We describe an information-theory-based measure of the quality of secondary structure prediction (RELINFO). RELINFO has a simple yet intuitive interpretation: it represents the factor by which secondary structure choice at a residue has been restricted by a prediction scheme. As an alternative interpretation of secondary structure prediction, RELINFO complements currently used methods by providing an information-based view as to why a prediction succeeds and fails. To demonstrate this score's capabilities, we applied RELINFO to an analysis of a large set of secondary structure predictions obtained from the first five rounds of the Critical Assessment of Structure Prediction (CASP) experiment. RELINFO is compared with two other common measures: percent correct (Q3) and secondary structure overlap (SOV). While the correlation between Q3 and RELINFO is approximately 0.85, RELINFO avoids certain disadvantages of Q3, including overestimating the quality of a prediction. The correlation between SOV and RELINFO is approximately 0.75. The valuable SOV measure unfortunately suffers from a saturation problem, and perhaps has unfairly given the general impression that secondary structure prediction has reached its limit since SOV hasn't improved much over the recent rounds of CASP. Although not a replacement for SOV, RELINFO has greater dispersion. Over the five rounds of CASP assessed here, RELINFO shows that predictions targets have been more difficult in successive CASP experiments, yet the predictions quality has continued to improve measurably over each round. In terms of information, the secondary structure prediction quality has almost doubled from CASP1 to CASP5. Therefore, as a different perspective of accuracy, RELINFO can help to improve prediction of protein secondary structure by providing a measure of difficulty as well as final quality of a prediction.

Recent successes of the energy landscape theory of protein folding and function

Protein folding and binding can be understood using energy landscape theory. When seeming deviations from the predictions of the funnel hypothesis are found, landscape theory helps us locate the cause. Sometimes the deviation reflects symmetry effects, allowing extra degeneracies to occur. Such effects seem to explain some kinetic anomalies in helical bundles. When binding processes were found to use apparently non-funneled landscapes this was traced to an inadequate understanding of biomolecular forces. The discrepancy allowed the discovery of new water-mediated forces - some of which act between hydrophilic residues. Introducing such forces into the algorithms greatly improves the quality of structure predictions.

Water in protein structure prediction.

Proteins have evolved to use water to help guide folding. A physically motivated, nonpairwise-additive model of water-mediated interactions added to a protein structure prediction Hamiltonian yields marked improvement in the quality of structure prediction for larger proteins. Free energy profile analysis suggests that long-range water-mediated potentials guide folding and smooth the underlying folding funnel. Analyzing simulation trajectories gives direct evidence that water-mediated interactions facilitate native-like packing of supersecondary structural elements. Long-range pairing of hydrophilic groups is an integral part of protein architecture. Specific water-mediated interactions are a universal feature of biomolecular recognition landscapes in both folding and binding.

Secondary structure prediction quality for naturally occurring amino acids in soluble proteins

To judge the performance of protein secondary structure prediction it is common to use performance measures that can report the prediction accuracy for each conformation of the three-state model ( α-helix, β-sheet and loop). Much more specific performance quality factors can be associated with each amino acid type found in each conformation. Such measures are introduced in this work and used to test both weak and strong features of secondary structure prediction with neural network algorithms. Proline in the loop conformation is the best predicted amino acid conformation. At the same time proline is the worst predicted amino acid in regular secondary structures. Other helix cap residues: glycine, serine, asparagine, aspartate and histidine are also poorly predicted in regular secondary structures. The overall percentage of correct predictions ranges from 77 for methionine to 65 for cysteine. Based on these results the prediction accuracy profile can be reported as a sequence of numbers along a polypeptide sequence for each protein tested with the chosen prediction scheme. Sequence segments associated with a low prediction accuracy will indicate that a training database of proteins was not adequate for the task of predicting such segments even by using the best available pattern recognition scheme.