Pau Bellot

NetBenchmark: a bioconductor package for reproducible benchmarks of gene regulatory network inference

BackgroundIn the last decade, a great number of methods for reconstructing gene regulatory networks from expression data have been proposed. However, very few tools and datasets allow to evaluate accurately and reproducibly those methods. Hence, we propose here a new tool, able to perform a systematic, yet fully reproducible, evaluation of transcriptional network inference methods.ResultsOur open-source and freely available Bioconductor package aggregates a large set of tools to assess the robustness of network inference algorithms against different simulators, topologies, sample sizes and noise intensities.ConclusionsThe benchmarking framework that uses various datasets highlights the specialization of some methods toward network types and data. As a result, it is possible to identify the techniques that have broad overall performances.

GENE EXPRESSION DATA CLASSIFICATION COMBINING HIERARCHICAL REPRESENTATION AND EFFICIENT FEATURE SELECTION

A general framework for microarray data classification is proposed in this paper. It produces precise and reliable classifiers through a two-step approach. At first, the original feature set is enhanced by a new set of features called metagenes. These new features are obtained through a hierarchical clustering process on the original data. Two different metagene generation rules have been analyzed, called Treelets clustering and Euclidean clustering. Metagenes creation is attractive for several reasons: first, they can improve the classification since they broaden the available feature space and capture the common behavior of similar genes reducing the residual measurement noise. Furthermore, by analyzing some of the chosen metagenes for classification with gene set enrichment analysis algorithms, it is shown how metagenes can summarize the behavior of functionally related probe sets. Additionally, metagenes can point out, still undocumented, highly discriminant probe sets numerically related to other probes endowed with prior biological information in order to contribute to the knowledge discovery process. The second step of the framework is the feature selection which applies the Improved Sequential Floating Forward Selection algorithm (IFFS) to properly choose a subset from the available feature set for classification composed of genes and metagenes. Considering the microarray sample scarcity problem, besides the classical error rate, a reliability measure is introduced to improve the feature selection process. Different scoring schemes are studied to choose the best one using both error rate and reliability. The Linear Discriminant Analysis classifier (LDA) has been used throughout this work, due to its good characteristics, but the proposed framework can be used with almost any classifier. The potential of the proposed framework has been evaluated analyzing all the publicly available datasets offered by the Micro Array Quality Control Study, phase II (MAQC). The comparative results showed that the proposed framework can compete with a wide variety of state of the art alternatives and it can obtain the best mean performance if a particular setup is chosen. A Monte Carlo simulation confirmed that the proposed framework obtains stable and repeatable results.

Likely effects of climate change on groundwater availability in a Mediterranean region of Southeastern Spain

Groundwater resources are typically the main fresh water source in arid and semi-arid regions. Natural recharge of aquifers is mainly based on precipitation; however, only heavy precipitation events (HPEs) are expected to produce appreciable aquifer recharge in these environments. In this work, we used daily precipitation and monthly water level time series from different locations over a Mediterranean region of Southeastern Spain to identify the critical threshold value to define HPEs that lead to appreciable aquifer recharge in this region. Wavelet and trend analyses were used to study the changes in the temporal distribution of the chosen HPEs (=20?mm?day-1) over the observed period 1953–2012 and its projected evolution by using 18 downscaled climate projections over the projected period 2040–2099. The used precipitation time series were grouped in 10 clusters according to similarities between them assessed by using Pearson correlations. Results showed that the critical HPE threshold for the study area is 20?mm?day-1. Wavelet analysis showed that observed significant seasonal and annual peaks in global wavelet spectrum in the first sub-period (1953–1982) are no longer significant in the second sub-period (1983–2012) in the major part of the ten clusters. This change is because of the reduction of the mean HPEs number, which showed a negative trend over the observed period in nine clusters and was significant in five of them. However, the mean size of HPEs showed a positive trend in six clusters. A similar tendency of change is expected over the projected period. The expected reduction of the mean HPEs number is two times higher under the high climate scenario (RCP8.5) than under the moderate scenario (RCP4.5). The mean size of these events is expected to increase under the two scenarios. The groundwater availability will be affected by the reduction of HPE number which will increase the length of no aquifer recharge periods (NARP) accentuating the groundwater drought in the region. Copyright

Study of Normalization and Aggregation Approaches for Consensus Network Estimation

Inferring gene regulatory networks from expression data is a very difficult problem that has raised the interest of the scientific community. Different algorithms have been proposed to try to solve this issue, but it has been shown that the different methods have some particular biases and strengths, and none of them is the best across all types of data and datasets. As a result, the idea of aggregating various network inferences through a consensus mechanism naturally arises. In this paper, a common framework to standardize already proposed consensus methods is presented, and based on this framework different proposals are introduced and analyzed in two different scenarios: Homogeneous and Heterogeneous. The first scenario reflects situations where the networks to be aggregated are rather similar because the are obtained with inference algorithms working on the same data, whereas the second scenario deals with very diverse networks because various sources of data are used to generate the individual networks. A procedure for combining multiple network inference algorithms is analyzed in a systematic way. The results show that there is a very significant difference between these two scenarios, and that the best way to combine networks in the Heterogeneous scenario is not the most commonly used. We show in particular that aggregation in the Heterogeneous scenario can be very beneficial if the individual networks are combined with our new proposed method Scale LSum.

Efficient combination of pairwise feature networks

This paper presents a novel method for the reconstruction of a neural network connectivity using calcium fluorescence data. We introduce a fast unsupervised method to integrate different networks that reconstructs structural connectivity from neuron activity. Our method improves the state-of-the-art reconstruction method General Transfer Entropy (GTE). We are able to better eliminate indirect links, improving therefore the quality of the network via a normalization and ensemble process of GTE and three new informative features. The approach is based on a simple combination of networks, which is remarkably fast. The performance of our approach is benchmarked on simulated time series provided at the connectomics challenge and also submitted at the public competition.

Study of Meta-analysis strategies for network inference using information-theoretic approaches

BackgroundReverse engineering of gene regulatory networks (GRNs) from gene expression data is a classical challenge in systems biology. Thanks to high-throughput technologies, a massive amount of gene-expression data has been accumulated in the public repositories. Modelling GRNs from multiple experiments (also called integrative analysis) has; therefore, naturally become a standard procedure in modern computational biology. Indeed, such analysis is usually more robust than the traditional approaches, which suffer from experimental biases and the low number of samples by analysing individual datasets.To date, there are mainly two strategies for the problem of interest: the first one (“data merging”) merges all datasets together and then infers a GRN whereas the other (“networks ensemble”) infers GRNs from every dataset separately and then aggregates them using some ensemble rules (such as ranksum or weightsum). Unfortunately, a thorough comparison of these two approaches is lacking.ResultsIn this work, we are going to present another meta-analysis approach for inferring GRNs from multiple studies. Our proposed meta-analysis approach, adapted to methods based on pairwise measures such as correlation or mutual information, consists of two steps: aggregating matrices of the pairwise measures from every dataset followed by extracting the network from the meta-matrix. Afterwards, we evaluate the performance of the two commonly used approaches mentioned above and our presented approach with a systematic set of experiments based on in silico benchmarks.ConclusionsWe proposed a first systematic evaluation of different strategies for reverse engineering GRNs from multiple datasets. Experiment results strongly suggest that assembling matrices of pairwise dependencies is a better strategy for network inference than the two commonly used ones.

Hierarchical clustering combining numerical and biological similarities for gene expression data classification

High throughput data analysis is a challenging problem due to the vast amount of available data. A major concern is to develop algorithms that provide accurate numerical predictions and biologically relevant results. A wide variety of tools exist in the literature using biological knowledge to evaluate analysis results. Only recently, some works have included biological knowledge inside the analysis process improving the prediction results. In this work, a knowledge integration scheme is proposed to improve the microarray classification results from [3]. Biological knowledge is used to infer biological similarity which is combined with the classical numerical similarity. The resulting similarity measure is used in a hierarchical clustering process producing new features called metagenes. The goal of the numerical and biological similarities integration is to produce metagenes involving more useful and significant gene signatures. The proposed algorithm has been tested on 7 publicly available datasets. The results have been compared with the state of the art method. The knowledge inclusion has proven beneficial both for the predictive ability, improving the results repeatability, and for the biological relevance after evaluating the produced signatures with two gene list analysis tools.

Ensemble learning and hierarchical data representation for microarray classification

The microarray data classification is an open and active research field. The development of more accurate algorithms is of great interest and many of the developed techniques can be straightforwardly applied in analyzing different kinds of omics data. In this work, an ensemble learning algorithm is applied within a classification framework that already got good predictive results. Ensemble techniques take individual experts, (i.e. classifiers), to combine them to improve the individual expert results with a voting scheme. In this case, a thinning algorithm is proposed which starts by using all the available experts and removes them one by one focusing on improving the ensemble vote. Two versions of a state of the art ensemble thinning algorithm have been tested and three key elements have been introduced to work with microarray data: the ensemble cohort definition, the nonexpert notion, which defines a set of excluded expert from the thinning process, and a rule to break ties in the thinning process. Experiments have been done on seven public datasets from the Microarray Quality Control study, MAQC. The proposed key elements have shown to be useful for the prediction performance and the studied ensemble technique shown to improve the state of the art results by producing classifiers with better predictions.

Multiclass cancer-microarray classification algorithm with pair-against-all redundancy

Multiclass cancer classification is still a challenging task in the field of machine learning. A novel multiclass approach is proposed in this work as a combination of multiple binary classifiers. It is an example of Error Correcting Output Codes algorithms, applying data transmission coding techniques to improve the classification as a combination of binary classifiers. The proposed method combines the One Against All, OAA, approach with a set of classifiers separating each class-pair from the rest, called Pair Against All, PAA. The OAA+PAA approach has been tested on seven publicly available datasets. It has been compared with the common OAA approach and with state of the art alternatives. The obtained results showed how the OAA+PAA algorithm consistently improves the OAA results, unlike other ECOC algorithms presented in the literature.

Microarray classification with hierarchical data representation and novel feature selection criteria

Microarray data classification is a challenging problem due to the high number of variables compared to the small number of available samples. An effective methodology to output a precise and reliable classifier is proposed in this work as an improvement of the algorithm in [1]. It considers the sample scarcity problem and the lack of data structure typical of microarrays. Both problem are assessed by a two-step approach applying hierarchical clustering to create new features called metagenes and introducing a novel feature ranking criterion, inside the wrapper feature selection task. The classification ability has been evaluated on 4 publicly available datasets from Micro Array Quality Control study phase II (MAQC) classified by 7 different endpoints. The global results have showed how the proposed approach obtains better prediction accuracy than a wide variety of state of the art alternatives.