RIP: the regulatory interaction predictor--a machine learning-based approach for predicting target genes of transcription factors

Recent Advances in the Prediction of Protein Structural Classes: Feature Descriptors and Machine Learning Algorithms

In the postgenomic age, rapid growth in the number of sequence-known proteins has been accompanied by much slower growth in the number of structure-known proteins (as a result of experimental limitations), and a widening gap between the two is evident. Because protein function is linked to protein structure, successful prediction of protein structure is of significant importance in protein function identification. Foreknowledge of protein structural class can help improve protein structure prediction with significant medical and pharmaceutical implications. Thus, a fast, suitable, reliable, and reasonable computational method for protein structural class prediction has become pivotal in bioinformatics. Here, we review recent efforts in protein structural class prediction from protein sequence, with particular attention paid to new feature descriptors, which extract information from protein sequence, and the use of machine learning algorithms in both feature selection and the construction of new classification models. These new feature descriptors include amino acid composition, sequence order, physicochemical properties, multiprofile Bayes, and secondary structure-based features. Machine learning methods, such as artificial neural networks (ANNs), support vector machine (SVM), K-nearest neighbor (KNN), random forest, deep learning, and examples of their application are discussed in detail. We also present our view on possible future directions, challenges, and opportunities for the applications of machine learning algorithms for prediction of protein structural classes.

Mixed Integer Linear Programming based machine learning approach identifies regulators of telomerase in yeast

Understanding telomere length maintenance mechanisms is central in cancer biology as their dysregulation is one of the hallmarks for immortalization of cancer cells. Important for this well-balanced control is the transcriptional regulation of the telomerase genes. We integrated Mixed Integer Linear Programming models into a comparative machine learning based approach to identify regulatory interactions that best explain the discrepancy of telomerase transcript levels in yeast mutants with deleted regulators showing aberrant telomere length, when compared to mutants with normal telomere length. We uncover novel regulators of telomerase expression, several of which affect histone levels or modifications. In particular, our results point to the transcription factors Sum1, Hst1 and Srb2 as being important for the regulation of EST1 transcription, and we validated the effect of Sum1 experimentally. We compiled our machine learning method leading to a user friendly package for R which can straightforwardly be applied to similar problems integrating gene regulator binding information and expression profiles of samples of e.g. different phenotypes, diseases or treatments.

Demystifying computer science for molecular ecologists

In this age of data-driven science and high-throughput biology, computational thinking is becoming an increasingly important skill for tackling both new and long-standing biological questions. However, despite its obvious importance and conspicuous integration into many areas of biology, computer science is still viewed as an obscure field that has, thus far, permeated into only a few of the biology curricula across the nation. A national survey has shown that lack of computational literacy in environmental sciences is the norm rather than the exception [Valle & Berdanier (2012) Bulletin of the Ecological Society of America, 93, 373-389]. In this article, we seek to introduce a few important concepts in computer science with the aim of providing a context-specific introduction aimed at research biologists. Our goal was to help biologists understand some of the most important mainstream computational concepts to better appreciate bioinformatics methods and trade-offs that are not obvious to the uninitiated.

Characterizing protein interactions employing a genome-wide siRNA cellular phenotyping screen

Characterizing the activating and inhibiting effect of protein-protein interactions (PPI) is fundamental to gain insight into the complex signaling system of a human cell. A plethora of methods has been suggested to infer PPI from data on a large scale, but none of them is able to characterize the effect of this interaction. Here, we present a novel computational development that employs mitotic phenotypes of a genome-wide RNAi knockdown screen and enables identifying the activating and inhibiting effects of PPIs. Exemplarily, we applied our technique to a knockdown screen of HeLa cells cultivated at standard conditions. Using a machine learning approach, we obtained high accuracy (82% AUC of the receiver operating characteristics) by cross-validation using 6,870 known activating and inhibiting PPIs as gold standard. We predicted de novo unknown activating and inhibiting effects for 1,954 PPIs in HeLa cells covering the ten major signaling pathways of the Kyoto Encyclopedia of Genes and Genomes, and made these predictions publicly available in a database. We finally demonstrate that the predicted effects can be used to cluster knockdown genes of similar biological processes in coherent subgroups. The characterization of the activating or inhibiting effect of individual PPIs opens up new perspectives for the interpretation of large datasets of PPIs and thus considerably increases the value of PPIs as an integrated resource for studying the detailed function of signaling pathways of the cellular system of interest.

Mathematical models which aim to describe cellular signaling start from constructing an interaction network of effectors, mediators and their effected target proteins. Several developments came up making it easier to put these links together. Besides tediously assembling knowledge from textbooks and research articles, experimental high-throughput methods were established like Yeast-2-Hybrid assays or Fluorescence Emission Resonance Transfer. However, these methods do not elucidate the effect of such interactions. We aimed inferring if an interaction in a specific cellular context is rather activating or inhibiting. We used cellular phenotypes of a genome-wide RNAi knockdown screen of live cells to identify such activating and inhibiting effects of protein interactions. The rationale behind it is that activating protein interactions should lead to similar phenotypes when their respective genes are knocked down, whereas an inhibiting protein interaction should lead to dissimilar phenotypes. Exemplarily, we applied our method to a phenotype screen of perturbed HeLa cells. Our predictions effectively reproduced textbook relationships between proteins or domains when comparing the predicted effects with pairs of effectors, receptors, kinases, phosphatases and of general signalling modules. The presented computational approach is generic and may enable elucidating the effects of studied interactions also of other cellular systems under more specific conditions.

Modulation of gene expression regulated by the transcription factor NF-κB/RelA

Modulators (Ms) are proteins that modify the activity of transcription factors (TFs) and influence expression of their target genes (TGs). To discover modulators of NF-κB/RelA, we first identified 365 NF-κB/RelA-binding proteins using liquid chromatography-tandem mass spectrometry (LC-MS/MS). We used a probabilistic model to infer 8349 (M, NF-κB/RelA, TG) triplets and their modes of modulatory action from our combined LC-MS/MS and ChIP-Seq (ChIP followed by next generation sequencing) data, published RelA modulators and TGs, and a compendium of gene expression profiles. Hierarchical clustering of the derived modulatory network revealed functional subnetworks and suggested new pathways modulating RelA transcriptional activity. The modulators with the highest number of TGs and most non-random distribution of action modes (measured by Shannon entropy) are consistent with published reports. Our results provide a repertoire of testable hypotheses for experimental validation. One of the NF-κB/RelA modulators we identified is STAT1. The inferred (STAT1, NF-κB/RelA, TG) triplets were validated by LC-selected reaction monitoring-MS and the results of STAT1 deletion in human fibrosarcoma cells. Overall, we have identified 562 NF-κB/RelA modulators, which are potential drug targets, and clarified mechanisms of achieving NF-κB/RelA multiple functions through modulators. Our approach can be readily applied to other TFs.

On protocols and measures for the validation of supervised methods for the inference of biological networks

Networks provide a natural representation of molecular biology knowledge, in particular to model relationships between biological entities such as genes, proteins, drugs, or diseases. Because of the effort, the cost, or the lack of the experiments necessary for the elucidation of these networks, computational approaches for network inference have been frequently investigated in the literature. In this paper, we examine the assessment of supervised network inference. Supervised inference is based on machine learning techniques that infer the network from a training sample of known interacting and possibly non-interacting entities and additional measurement data. While these methods are very effective, their reliable validation in silico poses a challenge, since both prediction and validation need to be performed on the basis of the same partially known network. Cross-validation techniques need to be specifically adapted to classification problems on pairs of objects. We perform a critical review and assessment of protocols and measures proposed in the literature and derive specific guidelines how to best exploit and evaluate machine learning techniques for network inference. Through theoretical considerations and in silico experiments, we analyze in depth how important factors influence the outcome of performance estimation. These factors include the amount of information available for the interacting entities, the sparsity and topology of biological networks, and the lack of experimentally verified non-interacting pairs.

The evolution of complex gene regulation by low-specificity binding sites

Requirements for gene regulation vary widely both within and among species. Some genes are constitutively expressed, whereas other genes require complex regulatory control. Transcriptional regulation is often controlled by a module of multiple transcription factor binding sites that, in combination, mediate the expression of a target gene. Here, we study how such regulatory modules evolve in response to natural selection. Using a population-genetic model, we show that complex regulatory modules which contain a larger number of binding sites must employ binding motifs that are less specific, on average, compared with smaller regulatory modules. This effect is extremely general, and it holds regardless of the selected binding logic that a module experiences. We attribute this phenomenon to the inability of stabilizing selection to maintain highly specific sites in large regulatory modules. Our analysis helps to explain broad empirical trends in the Saccharomyces cerevisiae regulatory network: those genes with a greater number of distinct transcriptional regulators feature less-specific binding motifs, compared with genes with fewer regulators. Our results also help to explain empirical trends in module size and motif specificity across species, ranging from prokaryotes to single-cellular and multi-cellular eukaryotes.

Passing messages between biological networks to refine predicted interactions

Regulatory network reconstruction is a fundamental problem in computational biology. There are significant limitations to such reconstruction using individual datasets, and increasingly people attempt to construct networks using multiple, independent datasets obtained from complementary sources, but methods for this integration are lacking. We developed PANDA (Passing Attributes between Networks for Data Assimilation), a message-passing model using multiple sources of information to predict regulatory relationships, and used it to integrate protein-protein interaction, gene expression, and sequence motif data to reconstruct genome-wide, condition-specific regulatory networks in yeast as a model. The resulting networks were not only more accurate than those produced using individual data sets and other existing methods, but they also captured information regarding specific biological mechanisms and pathways that were missed using other methodologies. PANDA is scalable to higher eukaryotes, applicable to specific tissue or cell type data and conceptually generalizable to include a variety of regulatory, interaction, expression, and other genome-scale data. An implementation of the PANDA algorithm is available at www.sourceforge.net/projects/panda-net.

Under-dominance constrains the evolution of negative autoregulation in diploids

Regulatory networks have evolved to allow gene expression to rapidly track changes in the environment as well as to buffer perturbations and maintain cellular homeostasis in the absence of change. Theoretical work and empirical investigation in Escherichia coli have shown that negative autoregulation confers both rapid response times and reduced intrinsic noise, which is reflected in the fact that almost half of Escherichia coli transcription factors are negatively autoregulated. However, negative autoregulation is rare amongst the transcription factors of Saccharomyces cerevisiae. This difference is surprising because E. coli and S. cerevisiae otherwise have similar profiles of network motifs. In this study we investigate regulatory interactions amongst the transcription factors of Drosophila melanogaster and humans, and show that they have a similar dearth of negative autoregulation to that seen in S. cerevisiae. We then present a model demonstrating that this stiking difference in the noise reduction strategies used amongst species can be explained by constraints on the evolution of negative autoregulation in diploids. We show that regulatory interactions between pairs of homologous genes within the same cell can lead to under-dominance — mutations which result in stronger autoregulation, and decrease noise in homozygotes, paradoxically can cause increased noise in heterozygotes. This severely limits a diploid's ability to evolve negative autoregulation as a noise reduction mechanism. Our work offers a simple and general explanation for a previously unexplained difference between the regulatory architectures of E. coli and yeast, Drosophila and humans. It also demonstrates that the effects of diploidy in gene networks can have counter-intuitive consequences that may profoundly influence the course of evolution.

All genes have to deal with intrinsic noise, and a variety of mechanisms have evolved to reduce it. One important mechanism of noise reduction for transcription factors is negative autoregulation, in which a gene product represses its own rate of transcription. Negative auotregulation occurs frequently in E. coli but, we find, occurs much more rarely in S. cerevisiae, D. melanogaster and humans. Whilst there are a great many important differences in the genetic architectures of these organisms, they tend to share, with the exception of negative autoregulation, similar profiles of network motifs. This makes the discrepancy in the degree of negative autoregulation all the more striking, as it lacks any obvious explanation. Our study presents a potential explanation, by comparing the evolvability of negative autoregulation as a noise reduction mechanism in haploids and diploids. We show that, in diploids, mutations that increase the strength of negative autoregulation at one gene copy often increase overall noise in gene expression. This results in under-dominance, in which heterozygotes are less fit than homozygotes. The result is that the evolution of negative autoregulation in diploids is significantly constrained. We verify our results using a combination of detailed molecular simulations and evolutionary simulations

Joint Bayesian inference of condition-specific miRNA and transcription factor activities from combined gene and microRNA expression data

MOTIVATION

There have been many successful experimental and bioinformatics efforts to elucidate transcription factor (TF)-target networks in several organisms. For many organisms, these annotations are complemented by miRNA-target networks of good quality. Attempts that use these networks in combination with gene expression data to draw conclusions on TF or miRNA activity are, however, still relatively sparse.

RESULTS

In this study, we propose Bayesian inference of regulation of transcriptional activity (BIRTA) as a novel approach to infer both, TF and miRNA activities, from combined miRNA and mRNA expression data in a condition specific way. That means our model explains mRNA and miRNA expression for a specific experimental condition by the activities of certain miRNAs and TFs, hence allowing for differentiating between switches from active to inactive (negative switch) and inactive to active (positive switch) forms. Extensive simulations of our model reveal its good prediction performance in comparison to other approaches. Furthermore, the utility of BIRTA is demonstrated at the example of Escherichia coli data comparing aerobic and anaerobic growth conditions, and by human expression data from pancreas and ovarian cancer.

AVAILABILITY AND IMPLEMENTATION

The method is implemented in the R package birta, which is freely available for Bio-conductor (>=2.10) on http://www.bioconductor.org/packages/release/bioc/html/birta.html.

A functional single-nucleotide polymorphism in the promoter of the gene encoding interleukin 6 is associated with susceptibility to tuberculosis

BACKGROUND

Genetic variation influences susceptibility or resistance to tuberculosis. Interleukin 6 (IL-6) contributes to protection against tuberculosis in mice. However, its role in regulating susceptibility or resistance to tuberculosis in humans is unclear.

METHODS

Genotyping of polymorphisms in IL-6 and IL-6R (CD126) genes was performed in 2 independent cohorts, an experimental population (495 cases and 358 controls) and a validation population (1383 cases and 1149 controls). The associations of the variants with tuberculosis were tested using 2 case-control association studies. In addition, the regulatory effects of single-nucleotide polymorphism rs1800796 (-572C > G) on IL-6 production in plasma and CD14(+) monocyte cultures stimulated with a Mycobacterium tuberculosis (M. tuberculosis) product were assessed.

RESULTS

The rs1800796 polymorphism is associated with increased resistance to tuberculosis (odds ratio [OR], 0.771; 95% confidential interval, .684-.870). The rs1800796GG genotype is strongly associated with reduced risk to tuberculosis (OR, 0.621; 95% CI, .460-.838). Interestingly, CD14(+) monocytes isolated from individuals with rs1800796GG genotype produced significantly less IL-6 in response to M. tuberculosis 19-kDa lipoprotein than those with CC or CG genotype.

CONCLUSIONS

We identified a genetic polymorphism in the IL-6 promoter that regulates cytokine production and host resistance to pulmonary tuberculosis in Chinese populations.