MOTIVATION: Signal transduction networks regulate many essential biological processes and are frequently aberrated in diseases such as cancer. A mechanistic understanding of such networks, and how they differ between cell populations, is essential to design effective treatment strategies. Typically, such networks are computationally reconstructed based on systematic perturbation experiments, followed by quantification of signaling protein activity. Recent technological advances now allow for the quantification of the activity of many (signaling) proteins simultaneously in single cells. This makes it feasible to reconstruct or quantify signaling networks without performing systematic perturbations. RESULTS: Here, we introduce single-cell modular response analysis (scMRA) and single-cell comparative network reconstruction (scCNR) to derive signal transduction networks by exploiting the heterogeneity of single-cell (phospho-)protein measurements. The methods treat stochastic variation in total protein abundances as natural perturbation experiments, whose effects propagate through the network and hence facilitate the reconstruction and quantification of the underlying signaling network. scCNR reconstructs cell population-specific networks, where cells from different populations have the same underlying topology, but the interaction strengths can differ between populations. We extensively validated scMRA and scCNR on simulated data, and applied it to unpublished data of (phospho-)protein measurements of EGFR-inhibitor-treated keratinocytes to recover signaling differences downstream of EGFR. scCNR will help to unravel the mechanistic signaling differences between cell populations, and will subsequently guide the development of well-informed treatment strategies. AVAILABILITY AND IMPLEMENTATION: The code used for scCNR in this study has been deposited on Zenodo https://doi.org/10.5281/zenodo.17600937 and is also available as a Python module at https://github.com/ibivu/scmra. Additionally, data and code to reproduce all figures is available at https://github.com/tstohn/scmra_analysis.

Abstract Motivation: Anti-cancer drugs may elicit resistance or sensitivity through mechanisms which involve several genomic layers. Nevertheless, we have demonstrated that gene expression contains most of the predictive capacity compared to the remaining omic data types. Unfortunately, this comes at a price: gene expression biomarkers are often hard to interpret and show poor robustness. Results: To capture the best of both worlds, i.e. the accuracy of gene expression and the robustness of other genomic levels, such as mutations, copy-number or methylation, we developed Percolate, a computational approach which extracts the joint signal between gene expression and the other omic data types. We developed an out-of-sample extension of Percolate which allows predictions on unseen samples without the necessity to recompute the joint signal on all data. We employed Percolate to extract the joint signal between gene expression and either mutations, copy-number or methylation, and used the out-of sample extension to perform response prediction on unseen samples. We showed that the joint signal recapitulates, and sometimes exceeds, the predictive performance achieved with each data type individually. Importantly, molecular signatures created by Percolate do not require gene expression to be evaluated, rendering them suitable to clinical applications where only one data type is available. Availability: Percolate is available as a Python 3.7 package and the scripts to reproduce the results are available here .

BACKGROUND: High-dimensional prediction considers data with more variables than samples. Generic research goals are to find the best predictor or to select variables. Results may be improved by exploiting prior information in the form of co-data, providing complementary data not on the samples, but on the variables. We consider adaptive ridge penalised generalised linear and Cox models, in which the variable-specific ridge penalties are adapted to the co-data to give a priori more weight to more important variables. The R-package ecpc originally accommodated various and possibly multiple co-data sources, including categorical co-data, i.e. groups of variables, and continuous co-data. Continuous co-data, however, were handled by adaptive discretisation, potentially inefficiently modelling and losing information. As continuous co-data such as external p values or correlations often arise in practice, more generic co-data models are needed. RESULTS: Here, we present an extension to the method and software for generic co-data models, particularly for continuous co-data. At the basis lies a classical linear regression model, regressing prior variance weights on the co-data. Co-data variables are then estimated with empirical Bayes moment estimation. After placing the estimation procedure in the classical regression framework, extension to generalised additive and shape constrained co-data models is straightforward. Besides, we show how ridge penalties may be transformed to elastic net penalties. In simulation studies we first compare various co-data models for continuous co-data from the extension to the original method. Secondly, we compare variable selection performance to other variable selection methods. The extension is faster than the original method and shows improved prediction and variable selection performance for non-linear co-data relations. Moreover, we demonstrate use of the package in several genomics examples throughout the paper. CONCLUSIONS: The R-package ecpc accommodates linear, generalised additive and shape constrained additive co-data models for the purpose of improved high-dimensional prediction and variable selection. The extended version of the package as presented here (version number 3.1.1 and higher) is available on ( https://cran.r-project.org/web/packages/ecpc/ ).

Motivation: Genome-wide measurements of genetic and epigenetic alterations are generating more and more high-dimensional binary data. The special mathematical characteristics of binary data make the direct use of the classical principal component analysis (PCA) model to explore low-dimensional structures less obvious. Although there are several PCA alternatives for binary data in the psychometric, data analysis and machine learning literature, they are not well known to the bioinformatics community. Results: In this article, we introduce the motivation and rationale of some parametric and nonparametric versions of PCA specifically geared for binary data. Using both realistic simulations of binary data as well as mutation, CNA and methylation data of the Genomic Determinants of Sensitivity in Cancer 1000 (GDSC1000), the methods were explored for their performance with respect to finding the correct number of components, overfit, finding back the correct low-dimensional structure, variable importance, etc. The results show that if a low-dimensional structure exists in the data, that most of the methods can find it. When assuming a probabilistic generating process is underlying the data, we recommend to use the parametric logistic PCA model, while when such an assumption is not valid and the data are considered as given, the nonparametric Gifi model is recommended. Availability: The codes to reproduce the results in this article are available at the homepage of the Biosystems Data Analysis group (www.bdagroup.nl).

MOTIVATION: Cell lines and patient-derived xenografts (PDXs) have been used extensively to understand the molecular underpinnings of cancer. While core biological processes are typically conserved, these models also show important differences compared to human tumors, hampering the translation of findings from pre-clinical models to the human setting. In particular, employing drug response predictors generated on data derived from pre-clinical models to predict patient response remains a challenging task. As very large drug response datasets have been collected for pre-clinical models, and patient drug response data are often lacking, there is an urgent need for methods that efficiently transfer drug response predictors from pre-clinical models to the human setting. RESULTS: We show that cell lines and PDXs share common characteristics and processes with human tumors. We quantify this similarity and show that a regression model cannot simply be trained on cell lines or PDXs and then applied on tumors. We developed PRECISE, a novel methodology based on domain adaptation that captures the common information shared amongst pre-clinical models and human tumors in a consensus representation. Employing this representation, we train predictors of drug response on pre-clinical data and apply these predictors to stratify human tumors. We show that the resulting domain-invariant predictors show a small reduction in predictive performance in the pre-clinical domain but, importantly, reliably recover known associations between independent biomarkers and their companion drugs on human tumors. AVAILABILITY AND IMPLEMENTATION: PRECISE and the scripts for running our experiments are available on our GitHub page (https://github.com/NKI-CCB/PRECISE). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Motivation: In biology, we are often faced with multiple datasets recorded on the same set of objects, such as multi-omics and phenotypic data of the same tumors. These datasets are typically not independent from each other. For example, methylation may influence gene expression, which may, in turn, influence drug response. Such relationships can strongly affect analyses performed on the data, as we have previously shown for the identification of biomarkers of drug response. Therefore, it is important to be able to chart the relationships between datasets. Results: We present iTOP, a methodology to infer a topology of relationships between datasets. We base this methodology on the RV coefficient, a measure of matrix correlation, which can be used to determine how much information is shared between two datasets. We extended the RV coefficient for partial matrix correlations, which allows the use of graph reconstruction algorithms, such as the PC algorithm, to infer the topologies. In addition, since multi-omics data often contain binary data (e.g. mutations), we also extended the RV coefficient for binary data. Applying iTOP to pharmacogenomics data, we found that gene expression acts as a mediator between most other datasets and drug response: only proteomics clearly shares information with drug response that is not present in gene expression. Based on this result, we used TANDEM, a method for drug response prediction, to identify which variables predictive of drug response were distinct to either gene expression or proteomics. Availability and implementation: An implementation of our methodology is available in the R package iTOP on CRAN. Additionally, an R Markdown document with code to reproduce all figures is provided as Supplementary Material. Supplementary information: Supplementary data are available at Bioinformatics online.

Motivation: Signal-transduction networks are often aberrated in cancer cells, and new anti-cancer drugs that specifically target oncogenes involved in signaling show great clinical promise. However, the effectiveness of such targeted treatments is often hampered by innate or acquired resistance due to feedbacks, crosstalks or network adaptations in response to drug treatment. A quantitative understanding of these signaling networks and how they differ between cells with different oncogenic mutations or between sensitive and resistant cells can help in addressing this problem. Results: Here, we present Comparative Network Reconstruction (CNR), a computational method to reconstruct signaling networks based on possibly incomplete perturbation data, and to identify which edges differ quantitatively between two or more signaling networks. Prior knowledge about network topology is not required but can straightforwardly be incorporated. We extensively tested our approach using simulated data and applied it to perturbation data from a BRAF mutant, PTPN11 KO cell line that developed resistance to BRAF inhibition. Comparing the reconstructed networks of sensitive and resistant cells suggests that the resistance mechanism involves re-establishing wild-type MAPK signaling, possibly through an alternative RAF-isoform. Availability and implementation: CNR is available as a python module at https://github.com/NKI-CCB/cnr. Additionally, code to reproduce all figures is available at https://github.com/NKI-CCB/CNR-analyses. Supplementary information: Supplementary data are available at Bioinformatics online.

Motivation: Intracellular signalling is realized by complex signalling networks, which are almost impossible to understand without network models, especially if feedbacks are involved. Modular Response Analysis (MRA) is a convenient modelling method to study signalling networks in various contexts. Results: We developed the software package STASNet (STeady-STate Analysis of Signalling Networks) that provides an augmented and extended version of MRA suited to model signalling networks from incomplete perturbation schemes and multi-perturbation data. Using data from the Dialogue on Reverse Engineering Assessment and Methods challenge, we show that predictions from STASNet models are among the top-performing methods. We applied the method to study the effect of SHP2, a protein that has been implicated in resistance to targeted therapy in colon cancer, using a novel dataset from the colon cancer cell line Widr and a SHP2-depleted derivative. We find that SHP2 is required for mitogen-activated protein kinase signalling, whereas AKT signalling only partially depends on SHP2. Availability and implementation: An R-package is available at https://github.com/molsysbio/STASNet. Supplementary information: Supplementary data are available at Bioinformatics online.

Motivation: Computational models in biology are frequently underdetermined, due to limits in our capacity to measure biological systems. In particular, mechanistic models often contain parameters whose values are not constrained by a single type of measurement. It may be possible to achieve better model determination by combining the information contained in different types of measurements. Bayesian statistics provides a convenient framework for this, allowing a quantification of the reduction in uncertainty with each additional measurement type. We wished to explore whether such integration is feasible and whether it can allow computational models to be more accurately determined. Results: We created an ordinary differential equation model of cell cycle regulation in budding yeast and integrated data from 13 different studies covering different experimental techniques. We found that for some parameters, a single type of measurement, relative time course mRNA expression, is sufficient to constrain them. Other parameters, however, were only constrained when two types of measurements were combined, namely relative time course and absolute transcript concentration. Comparing the estimates to measurements from three additional, independent studies, we found that the degradation and transcription rates indeed matched the model predictions in order of magnitude. The predicted translation rate was incorrect however, thus revealing a deficiency in the model. Since this parameter was not constrained by any of the measurement types separately, it was only possible to falsify the model when integrating multiple types of measurements. In conclusion, this study shows that integrating multiple measurement types can allow models to be more accurately determined. Availability and implementation: The models and files required for running the inference are included in the Supplementary information. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

Effective cancer treatment is crucially dependent on the identification of the biological processes that drive a tumor. However, multiple processes may be active simultaneously in a tumor. Clustering is inherently unsuitable to this task as it assigns a tumor to a single cluster. In addition, the wide availability of multiple data types per tumor provides the opportunity to profile the processes driving a tumor more comprehensively. Here we introduce Functional Sparse-Factor Analysis (funcSFA) to address these challenges. FuncSFA integrates multiple data types to define a lower dimensional space capturing the relevant variation. A tailor-made module associates biological processes with these factors. FuncSFA is inspired by iCluster, which we improve in several key aspects. First, we increase the convergence efficiency significantly, allowing the analysis of multiple molecular datasets that have not been pre-matched to contain only concordant features. Second, FuncSFA does not assign tumors to discrete clusters, but identifies the dominant driver processes active in each tumor. This is achieved by a regression of the factors on the RNA expression data followed by a functional enrichment analysis and manual curation step. We apply FuncSFA to the TCGA breast and lung datasets. We identify EMT and Immune processes common to both cancer types. In the breast cancer dataset we recover the known intrinsic subtypes and identify additional processes. These include immune infiltration and EMT, and processes driven by copy number gains on the 8q chromosome arm. In lung cancer we recover the major types (adenocarcinoma and squamous cell carcinoma) and processes active in both of these types. These include EMT, two immune processes, and the activity of the NFE2L2 transcription factor. We validate the breast cancer findings on the METABRIC set and demonstrate the translatability of the TCGA breast cancer factors to METABRIC. In summary, FuncSFA is a robust method to perform discovery of key driver processes in a collection of tumors through unsupervised integration of multiple molecular data types and functional annotation.

MOTIVATION: Clinical response to anti-cancer drugs varies between patients. A large portion of this variation can be explained by differences in molecular features, such as mutation status, copy number alterations, methylation and gene expression profiles. We show that the classic approach for combining these molecular features (Elastic Net regression on all molecular features simultaneously) results in models that are almost exclusively based on gene expression. The gene expression features selected by the classic approach are difficult to interpret as they often represent poorly studied combinations of genes, activated by aberrations in upstream signaling pathways. RESULTS: To utilize all data types in a more balanced way, we developed TANDEM, a two-stage approach in which the first stage explains response using upstream features (mutations, copy number, methylation and cancer type) and the second stage explains the remainder using downstream features (gene expression). Applying TANDEM to 934 cell lines profiled across 265 drugs (GDSC1000), we show that the resulting models are more interpretable, while retaining the same predictive performance as the classic approach. Using the more balanced contributions per data type as determined with TANDEM, we find that response to MAPK pathway inhibitors is largely predicted by mutation data, while predicting response to DNA damaging agents requires gene expression data, in particular SLFN11 expression. AVAILABILITY AND IMPLEMENTATION: TANDEM is available as an R package on CRAN (for more information, see http://ccb.nki.nl/software/tandem). CONTACT: [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Integrative network analysis methods provide robust interpretations of differential high-throughput molecular profile measurements. They are often used in a biomedical context-to generate novel hypotheses about the underlying cellular processes or to derive biomarkers for classification and subtyping. The underlying molecular profiles are frequently measured and validated on animal or cellular models. Therefore the results are not immediately transferable to human. In particular, this is also the case in a study of the recently discovered interleukin-17 producing helper T cells (Th17), which are fundamental for anti-microbial immunity but also known to contribute to autoimmune diseases. RESULTS: We propose a mathematical model for finding active subnetwork modules that are conserved between two species. These are sets of genes, one for each species, which (i) induce a connected subnetwork in a species-specific interaction network, (ii) show overall differential behavior and (iii) contain a large number of orthologous genes. We propose a flexible notion of conservation, which turns out to be crucial for the quality of the resulting modules in terms of biological interpretability. We propose an algorithm that finds provably optimal or near-optimal conserved active modules in our model. We apply our algorithm to understand the mechanisms underlying Th17 T cell differentiation in both mouse and human. As a main biological result, we find that the key regulation of Th17 differentiation is conserved between human and mouse. AVAILABILITY AND IMPLEMENTATION: xHeinz, an implementation of our algorithm, as well as all input data and results, are available at http://software.cwi.nl/xheinz and as a Galaxy service at http://services.cbib.u-bordeaux2.fr/galaxy in CBiB Tools. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Target enrichment, also referred to as DNA capture, provides an effective way to focus sequencing efforts on a genomic region of interest. Capture data are typically used to detect single-nucleotide variants. It can also be used to detect copy number alterations, which is particularly useful in the context of cancer, where such changes occur frequently. In copy number analysis, it is a common practice to determine log-ratios between test and control samples, but this approach results in a loss of information as it disregards the total coverage or intensity at a locus. RESULTS: We modeled the coverage or intensity of the test sample as a linear function of the control sample. This regression approach is able to deal with regions that are completely deleted, which are problematic for methods that use log-ratios. To demonstrate the utility of our approach, we used capture data to determine copy number for a set of 600 genes in a panel of nine breast cancer cell lines. We found high concordance between our results and those generated using a single-nucleotide polymorphsim genotyping platform. When we compared our results with other log-ratio-based methods, including ExomeCNV, we found that our approach produced better overall correlation with SNP data. AVAILABILITY: The algorithm is implemented in C and R and the code can be downloaded from http://bioinformatics.nki.nl/ocs/ CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

With the growth of genomics, research institutes are increasingly confronted with the task of providing computational support to biologists and the provision of computational facilities, both for laboratory-based bioinformaticians and bioinformatics research groups. The optimal organization of bioinformatics support units to meet these needs is a common topic of discussion and consideration. During the recent evaluation of a bioinformatics core facility, we ended up discussing—in general terms—what we consider basic guidelines for the organization of bioinformatics core units in large research institutes, based on the experience and developments in our own institutes. We thought that part of our discussion could be useful for others, keeping in mind that every organization has specific needs and requirements, and that bioinformatics support is a fast moving area that requires continuous adaptation. In the following, we summarized what we consider some key recommendations: Finally, there is always the very difficult issue regarding the optimal size of the bioinformatics support units. This depends on many factors such as the diversity of the topics being covered in the institute and the level of the computational skills of the biologists. We will, therefore, not give an exact figure, but rather provide figures from our own institutions which are very comparable in this regard. In our own institutions, NKI-AVL, CNIO and FIMM, we have approximately 1 full-time, institutionally funded bioinformatician to support 100 scientists. In addition, the research groups themselves have up to 5–10 times this number of bioinformaticians working in their own research groups. This number varies, of course, depending on the specific topics researched by a particular group. These so-called ‘embedded’ bioinfomatics researchers are involved in more research-oriented work and are largely funded by outside grants. In our opinion it is very important to develop structures that allow the ‘embedded’ bioinformaticians to meet and communicate. Bioinformatics departments have to clearly separate their service units from their research laboratories. This contributes to the transparency of the organization, especially regarding the allocation of funds. The organization of bioinformatics service unit should be clear with defined missions for each component. It is particularly important to separate tasks by topic. For example: support of database development and maintenance, statistical analysis of high-throughput data, automatic image acquisition and analysis of e.g. light microscopy data and analysis of next-generation sequencing data. The unit should install a ‘users committee’ that includes users and specialists of the unit. This committee would be responsible for establishing clear priorities on what technical platforms are supported or aborted, and defining the rules applied to prioritize projects for bioinformatics support. It is important that the services are provided in a highly transparent manner, including publicly accessible information on statistics of users, project progress, tools and use of resources. Particularly important is the interface between the bioinformatics unit and the medical activities in an institute. This requires the integration of biobanking, medical information (records) and the corresponding genomic information. This activity is key for translational medicine and it requires efficient, multi-disciplinary interaction between clinical and computational experts. This process can be greatly enhanced by a seamless integration of the relevant data streams. One of the key missions of the bioinformatics is to provide training to biologists at a basic level. It is important to also incorporate separate advanced training in new technical developments for ‘hybrid users’ or for expert bioinformaticians. This should preferably be done in collaboration with the bioinformatics research laboratories in the institute. It is useful to nominate a bioinformatics support person, whose task is to guide users in the use of public bioinformatics tools and databases as well as tools and methods developed within the institute. The goal is to facilitate the use of bioinformatics by the biologists using state of the art tools. In conclusion, we can state that we favor a model where a small core group of bioinformaticians provide transparently organized support on more general, institute-wide research problems, while the majority of the bioinformaticians are embedded in the research groups. Embedding ensures more direct interaction between the biologist and the bioinformatician, providing both researchers with a sense of ownership of the project. Not only does this elevate the skills level of both parties, but it also greatly enhances productivity.

MOTIVATION: We propose an efficient method to infer combinatorial association logic networks from multiple genome-wide measurements from the same sample. We demonstrate our method on a genetical genomics dataset, in which we search for Boolean combinations of multiple genetic loci that associate with transcript levels. RESULTS: Our method provably finds the global solution and is very efficient with runtimes of up to four orders of magnitude faster than the exhaustive search. This enables permutation procedures for determining accurate false positive rates and allows selection of the most parsimonious model. When applied to transcript levels measured in myeloid cells from 24 genotyped recombinant inbred mouse strains, we discovered that nine gene clusters are putatively modulated by a logical combination of trait loci rather than a single locus. A literature survey supports and further elucidates one of these findings. Due to our approach, optimal solutions for multi-locus logic models and accurate estimates of the associated false discovery rates become feasible. Our algorithm, therefore, offers a valuable alternative to approaches employing complex, albeit suboptimal optimization strategies to identify complex models. AVAILABILITY: The MATLAB code of the prototype implementation is available on: http://bioinformatics.tudelft.nl/ or http://bioinformatics.nki.nl/.

Tumorigenesis is a multi-step process in which normal cells transform into malignant tumors following the accumulation of genetic mutations that enable them to evade the growth control checkpoints that would normally suppress their growth or result in apoptosis. It is therefore important to identify those combinations of mutations that collaborate in cancer development and progression. DNA copy number alterations (CNAs) are one of the ways in which cancer genes are deregulated in tumor cells. We hypothesized that synergistic interactions between cancer genes might be identified by looking for regions of co-occurring gain and/or loss. To this end we developed a scoring framework to separate truly co-occurring aberrations from passenger mutations and dominant single signals present in the data. The resulting regions of high co-occurrence can be investigated for between-region functional interactions. Analysis of high-resolution DNA copy number data from a panel of 95 hematological tumor cell lines correctly identified co-occurring recombinations at the T-cell receptor and immunoglobulin loci in T- and B-cell malignancies, respectively, showing that we can recover truly co-occurring genomic alterations. In addition, our analysis revealed networks of co-occurring genomic losses and gains that are enriched for cancer genes. These networks are also highly enriched for functional relationships between genes. We further examine sub-networks of these networks, core networks, which contain many known cancer genes. The core network for co-occurring DNA losses we find seems to be independent of the canonical cancer genes within the network. Our findings suggest that large-scale, low-intensity copy number alterations may be an important feature of cancer development or maintenance by affecting gene dosage of a large interconnected network of functionally related genes.

MOTIVATION: Permutation tests have become a standard tool to assess the statistical significance of an event under investigation. The statistical significance, as expressed in a P-value, is calculated as the fraction of permutation values that are at least as extreme as the original statistic, which was derived from non-permuted data. This empirical method directly couples both the minimal obtainable P-value and the resolution of the P-value to the number of permutations. Thereby, it imposes upon itself the need for a very large number of permutations when small P-values are to be accurately estimated. This is computationally expensive and often infeasible. RESULTS: A method of computing P-values based on tail approximation is presented. The tail of the distribution of permutation values is approximated by a generalized Pareto distribution. A good fit and thus accurate P-value estimates can be obtained with a drastically reduced number of permutations when compared with the standard empirical way of computing P-values. AVAILABILITY: The Matlab code can be obtained from the corresponding author on request. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Spectra resulting from Surface-Enhanced Laser Desorption/Ionisation (SELDI) mass spectrometry measurements are constructed by combining sub-spectra, each of which are the result of a single firing of the laser responsible for the process of desorption/ionisation. These firings are performed at different locations of the spot on which the sample is analysed. The final spectrum is then constructed by summing over all these sub-spectra. This process is sub-optimal in that it can average out peaks from peptides that are present in low abundance or are unevenly distributed across the spot, particularly because the amount of noise varies considerably between sub-spectra. This argues for analysing sub-spectra separately and combining results afterwards. RESULTS: Here, we propose to analyse these sub-spectra one-by-one and combine the results using a framework which includes a significance test. This allows one to, for the first time, attach a confidence measure to detected peaks, based on the signal strength of a peak across sub-spectra. In a comparison with three other approaches the sub-spectral approach achieves a higher sensitivity and a low FDR. We further introduce the notion of peak-bags, which provide rich information about the sub-spectral contributions to a given peak. CONCLUSION: The proposed procedure offers better control over the process of distinguishing signal from noise, resulting in an improved performance over other available methods. Moreover, our method provides an implicit deconvolution of peaks, yielding insight in the actual shape of a peak, potentially aiding in a deeper understanding of peak distribution. AVAILABILITY: Implementations of the algorithm in R are available upon request.

BACKGROUND: Tumors have been hypothesized to be the result of a mixture of oncogenic events, some of which will be reflected in the gene expression of the tumor. Based on this hypothesis a variety of data-driven methods have been employed to decompose tumor expression profiles into component profiles, hypothetically linked to these events. Interpretation of the resulting data-driven components is often done by post-hoc comparison to, for instance, functional groupings of genes into gene sets. None of the data-driven methods allow the incorporation of that type of knowledge directly into the decomposition. RESULTS: We present a linear model which uses knowledge driven, pre-defined components to perform the decomposition. We solve this decomposition model in a constrained linear least squares fashion. From a variety of options, a lasso-based solution to the model performs best in linking single gene perturbation data to mouse data. Moreover, we show the decomposition of expression profiles from human breast cancer samples into single gene perturbation profiles and gene sets that are linked to the hallmarks of cancer. For these breast cancer samples we were able to discern several links between clinical parameters, and the decomposition weights, providing new insights into the biology of these tumors. Lastly, we show that the order in which the Lasso regularization shrinks the weights, unveils consensus patterns within clinical subgroups of the breast cancer samples. CONCLUSION: The proposed lasso-based constrained least squares decomposition provides a stable and relevant relation between samples and knowledge-based components, and is thus a viable alternative to data-driven methods. In addition, the consensus order of component importance within clinical subgroups provides a better molecular characterization of the subtypes.

MOTIVATION: Cells receive a wide variety of environmental signals, which are often processed combinatorially to generate specific genetic responses. Changes in transcript levels, as observed across different environmental conditions, can, to a large extent, be attributed to changes in the activity of transcription factors (TFs). However, in unraveling these transcription regulation networks, the actual environmental signals are often not incorporated into the model, simply because they have not been measured. The unquantified heterogeneity of the environmental parameters across microarray experiments frustrates regulatory network inference. RESULTS: We propose an inference algorithm that models the influence of environmental parameters on gene expression. The approach is based on a yeast microarray compendium of chemostat steady-state experiments. Chemostat cultivation enables the accurate control and measurement of many of the key cultivation parameters, such as nutrient concentrations, growth rate and temperature. The observed transcript levels are explained by inferring the activity of TFs in response to combinations of cultivation parameters. The interplay between activated enhancers and repressors that bind a gene promoter determine the possible up- or downregulation of the gene. The model is translated into a linear integer optimization problem. The resulting regulatory network identifies the combinatorial effects of environmental parameters on TF activity and gene expression. AVAILABILITY: The Matlab code is available from the authors upon request. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Mass spectrometry for biological data analysis is an active field of research, providing an efficient way of high-throughput proteome screening. A popular variant of mass spectrometry is SELDI, which is often used to measure sample populations with the goal of developing (clinical) classifiers. Unfortunately, not only is the data resulting from such measurements quite noisy, variance between replicate measurements of the same sample can be high as well. Normalisation of spectra can greatly reduce the effect of this technical variance and further improve the quality and interpretability of the data. However, it is unclear which normalisation method yields the most informative result. RESULTS: In this paper, we describe the first systematic comparison of a wide range of normalisation methods, using two objectives that should be met by a good method. These objectives are minimisation of inter-spectra variance and maximisation of signal with respect to class separation. The former is assessed using an estimation of the coefficient of variation, the latter using the classification performance of three types of classifiers on real-world datasets representing two-class diagnostic problems. To obtain a maximally robust evaluation of a normalisation method, both objectives are evaluated over multiple datasets and multiple configurations of baseline correction and peak detection methods. Results are assessed for statistical significance and visualised to reveal the performance of each normalisation method, in particular with respect to using no normalisation. The normalisation methods described have been implemented in the freely available MASDA R-package. CONCLUSION: In the general case, normalisation of mass spectra is beneficial to the quality of data. The majority of methods we compared performed significantly better than the case in which no normalisation was used. We have shown that normalisation methods that scale spectra by a factor based on the dispersion (e.g., standard deviation) of the data clearly outperform those where a factor based on the central location (e.g., mean) is used. Additional improvements in performance are obtained when these factors are estimated locally, using a sliding window within spectra, instead of globally, over full spectra. The underperforming category of methods using a globally estimated factor based on the central location of the data includes the method used by the majority of SELDI users.

BACKGROUND: Array comparative genome hybridization (aCGH) provides information about genomic aberrations. Alterations in the DNA copy number may cause the cell to malfunction, leading to cancer. Therefore, the identification of DNA amplifications or deletions across tumors may reveal key genes involved in cancer and improve our understanding of the underlying biological processes associated with the disease. RESULTS: We propose a supervised algorithm for the analysis of aCGH data and the identification of regions of chromosomal alteration (SIRAC). We first determine the DNA-probes that are important to distinguish the classes of interest, and then evaluate in a systematic and robust scheme if these relevant DNA-probes are closely located, i.e. form a region of amplification/deletion. SIRAC does not need any preprocessing of the aCGH datasets, and requires only few, intuitive parameters. CONCLUSION: We illustrate the features of the algorithm with the use of a simple artificial dataset. The results on two breast cancer datasets show promising outcomes that are in agreement with previous findings, but SIRAC better pinpoints the dissimilarities between the classes of interest.

BACKGROUND: Gene selection is an important step when building predictors of disease state based on gene expression data. Gene selection generally improves performance and identifies a relevant subset of genes. Many univariate and multivariate gene selection approaches have been proposed. Frequently the claim is made that genes are co-regulated (due to pathway dependencies) and that multivariate approaches are therefore per definition more desirable than univariate selection approaches. Based on the published performances of all these approaches a fair comparison of the available results can not be made. This mainly stems from two factors. First, the results are often biased, since the validation set is in one way or another involved in training the predictor, resulting in optimistically biased performance estimates. Second, the published results are often based on a small number of relatively simple datasets. Consequently no generally applicable conclusions can be drawn. RESULTS: In this study we adopted an unbiased protocol to perform a fair comparison of frequently used multivariate and univariate gene selection techniques, in combination with a ränge of classifiers. Our conclusions are based on seven gene expression datasets, across several cancer types. CONCLUSION: Our experiments illustrate that, contrary to several previous studies, in five of the seven datasets univariate selection approaches yield consistently better results than multivariate approaches. The simplest multivariate selection approach, the Top Scoring method, achieves the best results on the remaining two datasets. We conclude that the correlation structures, if present, are difficult to extract due to the small number of samples, and that consequently, overly-complex gene selection algorithms that attempt to extract these structures are prone to overtraining.

Retroviral insertional mutagenesis screens, which identify genes involved in tumor development in mice, have yielded a substantial number of retroviral integration sites, and this number is expected to grow substantially due to the introduction of high-throughput screening techniques. The data of various retroviral insertional mutagenesis screens are compiled in the publicly available Retroviral Tagged Cancer Gene Database (RTCGD). Integrally analyzing these screens for the presence of common insertion sites (CISs, i.e., regions in the genome that have been hit by viral insertions in multiple independent tumors significantly more than expected by chance) requires an approach that corrects for the increased probability of finding false CISs as the amount of available data increases. Moreover, significance estimates of CISs should be established taking into account both the noise, arising from the random nature of the insertion process, as well as the bias, stemming from preferential insertion sites present in the genome and the data retrieval methodology. We introduce a framework, the kernel convolution (KC) framework, to find CISs in a noisy and biased environment using a predefined significance level while controlling the family-wise error (FWE) (the probability of detecting false CISs). Where previous methods use one, two, or three predetermined fixed scales, our method is capable of operating at any biologically relevant scale. This creates the possibility to analyze the CISs in a scale space by varying the width of the CISs, providing new insights in the behavior of CISs across multiple scales. Our method also features the possibility of including models for background bias. Using simulated data, we evaluate the KC framework using three kernel functions, the Gaussian, triangular, and rectangular kernel function. We applied the Gaussian KC to the data from the combined set of screens in the RTCGD and found that 53% of the CISs do not reach the significance threshold in this combined setting. Still, with the FWE under control, application of our method resulted in the discovery of eight novel CISs, which each have a probability less than 5% of being false detections.

MOTIVATION: Microarray gene expression data are increasingly employed to identify sets of marker genes that accurately predict disease development and outcome in cancer. Many computational approaches have been proposed to construct such predictors. However, there is, as yet, no objective way to evaluate whether a new approach truly improves on the current state of the art. In addition no 'standard' computational approach has emerged which enables robust outcome prediction. RESULTS: An important contribution of this work is the description of a principled training and validation protocol, which allows objective evaluation of the complete methodology for constructing a predictor. We review the possible choices of computational approaches, with specific emphasis on predictor choice and reporter selection strategies. Employing this training-validation protocol, we evaluated different reporter selection strategies and predictors on six gene expression datasets of varying degrees of difficulty. We demonstrate that simple reporter selection strategies (forward filtering and shrunken centroids) work surprisingly well and outperform partial least squares in four of the six datasets. Similarly, simple predictors, such as the nearest mean classifier, outperform more complex classifiers. Our training-validation protocol provides a robust methodology to evaluate the performance of new computational approaches and to objectively compare outcome predictions on different datasets.

The training of neural net classifiers is often hampered by the occurrence of local minima, which results in the attainment of inferior classification performance. It has been shown that the occurrence of local minima in the criterion function is often related to specific patterns of defects in the classifier. In particular, three main causes for local minima were identified. Such an understanding of the physical correlates of local minima suggests sensible ways of choosing the weights from which the training process is initiated. A method of initialization is introduced and shown to decrease the possibility of local minima occurring on various test problems.