Reproducible cancer biomarker discovery in SELDI-TOF MS using different pre-processing algorithms

Background

There has been much interest in differentiating diseased and normal samples using biomarkers derived from mass spectrometry (MS) studies. However, biomarker identification for specific diseases has been hindered by irreproducibility. Specifically, a peak profile extracted from a dataset for biomarker identification depends on a data pre-processing algorithm. Until now, no widely accepted agreement has been reached.

Results

In this paper, we investigated the consistency of biomarker identification using differentially expressed (DE) peaks from peak profiles produced by three widely used average spectrum-dependent pre-processing algorithms based on SELDI-TOF MS data for prostate and breast cancers. Our results revealed two important factors that affect the consistency of DE peak identification using different algorithms. One factor is that some DE peaks selected from one peak profile were not detected as peaks in other profiles, and the second factor is that the statistical power of identifying DE peaks in large peak profiles with many peaks may be low due to the large scale of the tests and small number of samples. Furthermore, we demonstrated that the DE peak detection power in large profiles could be improved by the stratified false discovery rate (FDR) control approach and that the reproducibility of DE peak detection could thereby be increased.

Conclusions

Comparing and evaluating pre-processing algorithms in terms of reproducibility can elucidate the relationship among different algorithms and also help in selecting a pre-processing algorithm. The DE peaks selected from small peak profiles with few peaks for a dataset tend to be reproducibly detected in large peak profiles, which suggests that a suitable pre-processing algorithm should be able to produce peaks sufficient for identifying useful and reproducible biomarkers.     

PCAS – a precomputed proteome annotation database resource

Background  

Many model proteomes or "complete" sets of proteins of given organisms are now publicly available. Much effort has been invested in computational annotation of those "draft" proteomes. Motif or domain based algorithms play a pivotal role in functional classification of proteins. Employing most available computational algorithms, mainly motif or domain recognition algorithms, we set up to develop an online proteome annotation system with integrated proteome annotation data to complement existing resources.     

FITBAR: a web tool for the robust prediction of prokaryotic regulons

Background  

The binding of regulatory proteins to their specific DNA targets determines the accurate expression of the neighboring genes. The in silico prediction of new binding sites in completely sequenced genomes is a key aspect in the deeper understanding of gene regulatory networks. Several algorithms have been described to discriminate against false-positives in the prediction of new binding targets; however none of them has been implemented so far to assist the detection of binding sites at the genomic scale.     

ANMM4CBR: a case-based reasoning method for gene expression data classification

Background  

Accurate classification of microarray data is critical for successful clinical diagnosis and treatment. The "curse of dimensionality" problem and noise in the data, however, undermines the performance of many algorithms.     

Partition function and base pairing probabilities of RNA heterodimers

Background  

RNA has been recognized as a key player in cellular regulation in recent years. In many cases, non-coding RNAs exert their function by binding to other nucleic acids, as in the case of microRNAs and snoRNAs. The specificity of these interactions derives from the stability of inter-molecular base pairing. The accurate computational treatment of RNA-RNA binding therefore lies at the heart of target prediction algorithms.     

GC/MS based metabolomics: development of a data mining system for metabolite identification by using soft independent modeling of class analogy (SIMCA)

Background  

The goal of metabolomics analyses is a comprehensive and systematic understanding of all metabolites in biological samples. Many useful platforms have been developed to achieve this goal. Gas chromatography coupled to mass spectrometry (GC/MS) is a well-established analytical method in metabolomics study, and 200 to 500 peaks are routinely observed with one biological sample. However, only ~100 metabolites can be identified, and the remaining peaks are left as "unknowns".     

The accuracy of several multiple sequence alignment programs for proteins

Background  

There have been many algorithms and software programs implemented for the inference of multiple sequence alignments of protein and DNA sequences. The "true" alignment is usually unknown due to the incomplete knowledge of the evolutionary history of the sequences, making it difficult to gauge the relative accuracy of the programs.     

An effective approach for identification of <Emphasis Type="Italic">in vivo</Emphasis> protein-DNA binding sites from paired-end ChIP-Seq data

Background  

ChIP-Seq, which combines chromatin immunoprecipitation (ChIP) with high-throughput massively parallel sequencing, is increasingly being used for identification of protein-DNA interactions in vivo in the genome. However, to maximize the effectiveness of data analysis of such sequences requires the development of new algorithms that are able to accurately predict DNA-protein binding sites.     

Quantitative prediction of mouse class I MHC peptide binding affinity using support vector machine regression (SVR) models

Background  

The binding between peptide epitopes and major histocompatibility complex proteins (MHCs) is an important event in the cellular immune response. Accurate prediction of the binding between short peptides and the MHC molecules has long been a principal challenge for immunoinformatics. Recently, the modeling of MHC-peptide binding has come to emphasize quantitative predictions: instead of categorizing peptides as "binders" or "non-binders" or as "strong binders" and "weak binders", recent methods seek to make predictions about precise binding affinities.     

Prediction of protein binding sites in protein structures using hidden Markov support vector machine

Background  

Predicting the binding sites between two interacting proteins provides important clues to the function of a protein. Recent research on protein binding site prediction has been mainly based on widely known machine learning techniques, such as artificial neural networks, support vector machines, conditional random field, etc. However, the prediction performance is still too low to be used in practice. It is necessary to explore new algorithms, theories and features to further improve the performance.     

ASPDock: protein-protein docking algorithm using atomic solvation parameters model

Background  

Atomic Solvation Parameters (ASP) model has been proven to be a very successful method of calculating the binding free energy of protein complexes. This suggests that incorporating it into docking algorithms should improve the accuracy of prediction. In this paper we propose an FFT-based algorithm to calculate ASP scores of protein complexes and develop an ASP-based protein-protein docking method (ASPDock).     

An aggregate analysis of many predicted structures to reduce errors in protein structure comparison caused by conformational flexibility

Background

Conformational flexibility creates errors in the comparison of protein structures. Even small changes in backbone or sidechain conformation can radically alter the shape of ligand binding cavities. These changes can cause structure comparison programs to overlook functionally related proteins with remote evolutionary similarities, and cause others to incorrectly conclude that closely related proteins have different binding preferences, when their specificities are actually similar. Towards the latter effort, this paper applies protein structure prediction algorithms to enhance the classification of homologous proteins according to their binding preferences, despite radical conformational differences.

Methods

Specifically, structure prediction algorithms can be used to "remodel" existing structures against the same template. This process can return proteins in very different conformations to similar, objectively comparable states. Operating on close homologs exploits the accuracy of structure predictions on closely related proteins, but structure prediction is often a nondeterministic process. Identical inputs can generate subtly different models with very different binding cavities that make structure comparison difficult. We present a first method to mitigate such errors, called "medial remodeling", that examines a large number of predicted structures to eliminate extreme models of the same binding cavity.

Results

Our results, on the enolase and tyrosine kinase superfamilies, demonstrate that remodeling can enable proteins in very different conformations to be returned to states that can be objectively compared. Structures that would have been erroneously classified as having different binding preferences were often correctly classified after remodeling, while structures that would have been correctly classified as having different binding preferences almost always remained distinct. The enolase superfamily, which exhibited less sequential diversity than the tyrosine kinase superfamily, was classified more accurately after remodeling than the tyrosine kinases. Medial remodeling reduced errors from models with unusual perturbations that distort the shape of the binding site, enhancing classification accuracy.

Conclusions

This paper demonstrates that protein structure prediction can compensate for conformational variety in the comparison of protein-ligand binding sites. While protein structure prediction introduces new uncertainties into the structure comparison problem, our results indicate that unusual models can be ignored through an analysis of many models, using techniques like medial remodeling. These results point to applications of protein structure comparison that extend beyond existing crystal structures.

     

Bayesian estimation of the number of protonation sites for urinary metabolites from NMR spectroscopic data

Introduction

To aid the development of better algorithms for \(^1\)H NMR data analysis, such as alignment or peak-fitting, it is important to characterise and model chemical shift changes caused by variation in pH. The number of protonation sites, a key parameter in the theoretical relationship between pH and chemical shift, is traditionally estimated from the molecular structure, which is often unknown in untargeted metabolomics applications.

Objective

We aim to use observed NMR chemical shift titration data to estimate the number of protonation sites for a range of urinary metabolites.

Methods

A pool of urine from healthy subjects was titrated in the range pH 2–12, standard \(^1\)H NMR spectra were acquired and positions of 51 peaks (corresponding to 32 identified metabolites) were recorded. A theoretical model of chemical shift was fit to the data using a Bayesian statistical framework, using model selection procedures in a Markov Chain Monte Carlo algorithm to estimate the number of protonation sites for each molecule.

Results

The estimated number of protonation sites was found to be correct for 41 out of 51 peaks. In some cases, the number of sites was incorrectly estimated, due to very close pKa values or a limited amount of data in the required pH range.

Conclusions

Given appropriate data, it is possible to estimate the number of protonation sites for many metabolites typically observed in \(^1\)H NMR metabolomics without knowledge of the molecular structure. This approach may be a valuable resource for the development of future automated metabolite alignment, annotation and peak fitting algorithms.