Large Language Models (LLMs) have demonstrated remarkable capabilities across a variety of domains. However, issues such as hallucinations, implicit extrapolations, and inappropriate interpretations currently limit their wide adoption in scientific research. Retrieval-augmented Generation (RAG) has emerged as a promising technique to enhance the accuracy and reliability of LLM outputs. RAG integrates external information with LLM-generated content, ensuring more dependable results and enabling the inclusion of proprietary or domain-specific knowledge unavailable during the initial LLM training. In this work, we present an extensible LLM RAG architecture designed to answer complex scientific questions that can reference user-provided experimental data. The system uses domain-specific knowledge (RAGdom) to respond to general inquiries, supporting answers with citations from authoritative sources. These responses can be recursively integrated to provide user directed context for experiments (RAGexp), allowing subsequent questions to utilize both general and experiment-specific information. Our system can be easily extended with user-defined domain-specific modules for functionally interrogating quantitative experimental results (RAGfun). We demonstrate its application in analyzing, summarizing, interpreting, and assisting with synthesizing results from a gene expression comparison of two phenotypes. The RAGdomcomponent encompasses biological pathways, genes, and disease-related knowledge. The RAGexpcomponent is derived from user queries of differentially expressed gene (DEG) and pathway enrichment analysis (PEA) results. Additionally, RAGfunmodules are designed to interface with DEG and PEA data, enabling quantitative experimental results to be integrated seamlessly into user questions. Unlike traditional LLM outputs, our system’s results are free of hallucinations and include paragraph-level citations to the supporting literature. We compared the quality of our system’s output with those generated by LLM-only models such as GPT-3, GPT-4, Llama2, PaLM2, and BioGPT, and found our approach to be superior.

MOTIVATION: Large Language Models (LLMs) have provided spectacular results across a wide variety of domains. However, persistent concerns about hallucination and fabrication of authoritative sources raise serious issues for their integral use in scientific research. Retrieval-augmented generation (RAG) is a technique for making data and documents, otherwise unavailable during training, available to the LLM for reasoning tasks. In addition to making dynamic and quantitative data available to the LLM, RAG provides the means by which to carefully control and trace source material, thereby ensuring results are accurate, complete, and authoritative. RESULTS: Here, we introduce LmRaC, an LLM-based tool capable of answering complex scientific questions in the context of a user's own experimental results. LmRaC allows users to dynamically build domain specific knowledge-bases from PubMed sources (RAGdom). Answers are drawn solely from this RAG with citations to the paragraph level, virtually eliminating any chance of hallucination or fabrication. These answers can then be used to construct an experimental context (RAGexp) that, along with user supplied documents (e.g. design, protocols) and quantitative results, can be used to answer questions about the user's specific experiment. Questions about quantitative experimental data are integral to LmRaC and are supported by a user-defined and functionally extensible REST API server (RAGfun). AVAILABILITY AND IMPLEMENTATION: Detailed documentation for LmRaC along with a sample REST API server for defining user functions can be found at https://github.com/dbcraig/LmRaC. The LmRaC web application image can be pulled from Docker Hub (https://hub.docker.com) as dbcraig/lmrac.

MOTIVATION: COVID-19 has several distinct clinical phases: a viral replication phase, an inflammatory phase and in some patients, a hyper-inflammatory phase. High mortality is associated with patients developing cytokine storm syndrome. Treatment of hyper-inflammation in these patients using existing approved therapies with proven safety profiles could address the immediate need to reduce mortality. RESULTS: We analyzed the changes in the gene expression, pathways and putative mechanisms induced by SARS-CoV2 in NHBE, and A549 cells, as well as COVID-19 lung versus their respective controls. We used these changes to identify FDA approved drugs that could be repurposed to help COVID-19 patients with severe symptoms related to hyper-inflammation. We identified methylprednisolone (MP) as a potential leading therapy. The results were then confirmed in five independent validation datasets including Vero E6 cells, lung and intestinal organoids, as well as additional patient lung sample versus their respective controls. Finally, the efficacy of MP was validated in an independent clinical study. Thirty-day all-cause mortality occurred at a significantly lower rate in the MP-treated group compared to control group (29.6% versus 16.6%, P = 0.027). Clinical results confirmed the in silico prediction that MP could improve outcomes in severe cases of COVID-19. A low number needed to treat (NNT = 5) suggests MP may be more efficacious than dexamethasone or hydrocortisone. AVAILABILITY AND IMPLEMENTATION: iPathwayGuide is available at https://advaitabio.com/ipathwayguide/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Recent advances in biomedical research have made massive amount of transcriptomic data available in public repositories from different sources. Due to the heterogeneity present in the individual experiments, identifying reproducible biomarkers for a given disease from multiple independent studies has become a major challenge. The widely used meta-analysis approaches, such as Fisher's method, Stouffer's method, minP and maxP, have at least two major limitations: (i) they are sensitive to outliers, and (ii) they perform only one statistical test for each individual study, and hence do not fully utilize the potential sample size to gain statistical power. RESULTS: Here, we propose a gene-level meta-analysis framework that overcomes these limitations and identifies a gene signature that is reliable and reproducible across multiple independent studies of a given disease. The approach provides a comprehensive global signature that can be used to understand the underlying biological phenomena, and a smaller test signature that can be used to classify future samples of a given disease. We demonstrate the utility of the framework by constructing disease signatures for influenza and Alzheimer's disease using nine datasets including 1108 individuals. These signatures are then validated on 12 independent datasets including 912 individuals. The results indicate that the proposed approach performs better than the majority of the existing meta-analysis approaches in terms of both sensitivity as well as specificity. The proposed signatures could be further used in diagnosis, prognosis and identification of therapeutic targets. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

SUMMARY: Since cancer is a heterogeneous disease, tumor subtyping is crucial for improved treatment and prognosis. We have developed a subtype discovery tool, called PINSPlus, that is: (i) robust against noise and unstable quantitative assays, (ii) able to integrate multiple types of omics data in a single analysis and (iii) dramatically superior to established approaches in identifying known subtypes and novel subgroups with significant survival differences. Our validation on 12,158 samples from 44 datasets shows that PINSPlus vastly outperforms other approaches. The software is easy-to-use and can partition hundreds of patients in a few minutes on a personal computer. AVAILABILITY AND IMPLEMENTATION: The package is available at https://cran.r-project.org/package=PINSPlus. Data and R script used in this manuscript are available at https://bioinformatics.cse.unr.edu/software/PINSPlus/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Drug repurposing is a potential alternative to the classical drug discovery pipeline. Repurposing involves finding novel indications for already approved drugs. In this work, we present a novel machine learning-based method for drug repurposing. This method explores the anti-similarity between drugs and a disease to uncover new uses for the drugs. More specifically, our proposed method takes into account three sources of information: (i) large-scale gene expression profiles corresponding to human cell lines treated with small molecules, (ii) gene expression profile of a human disease and (iii) the known relationship between Food and Drug Administration (FDA)-approved drugs and diseases. Using these data, our proposed method learns a similarity metric through a supervised machine learning-based algorithm such that a disease and its associated FDA-approved drugs have smaller distance than the other disease-drug pairs. RESULTS: We validated our framework by showing that the proposed method incorporating distance metric learning technique can retrieve FDA-approved drugs for their approved indications. Once validated, we used our approach to identify a few strong candidates for repurposing. AVAILABILITY AND IMPLEMENTATION: The R scripts are available on demand from the authors. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

DNA methylation is an important epigenetic mechanism that plays a crucial role in cellular regulatory systems. Recent advancements in sequencing technologies now enable us to generate high-throughput methylation data and to measure methylation up to single-base resolution. This wealth of data does not come without challenges, and one of the key challenges in DNA methylation studies is to identify the significant differences in the methylation levels of the base pairs across distinct biological conditions. Several computational methods have been developed to identify differential methylation using bisulfite sequencing data; however, there is no clear consensus among existing approaches. A comprehensive survey of these approaches would be of great benefit to potential users and researchers to get a complete picture of the available resources. In this article, we present a detailed survey of 22 such approaches focusing on their underlying statistical models, primary features, key advantages and major limitations. Importantly, the intrinsic drawbacks of the approaches pointed out in this survey could potentially be addressed by future research.

Motivation: Epigenetic mechanisms are known to play a major role in breast cancer. However, the role of 5-hydroxymethylcytosine (5hmC) remains understudied. We hypothesize that 5hmC mediates redox regulation of gene expression in an aggressive subtype known as triple negative breast cancer (TNBC). To address this, our objective was to highlight genes that may be the target of this process by identifying redox-regulated, antioxidant-sensitive, gene-localized 5hmC changes associated with mRNA changes in TNBC cells. Results: We proceeded to develop an approach to integrate novel Pvu-sequencing and RNA-sequencing data. The result of our approach to merge genome-wide, high-throughput TNBC cell line datasets to identify significant, concordant 5hmC and mRNA changes in response to antioxidant treatment produced a gene set with relevance to cancer stem cell function. Moreover, we have established a method that will be useful for continued research of 5hmC in TNBC cells and tissue samples. Availability and implementation: Data are available at Gene Expression Omnibus (GEO) under accession number GSE103850. Contact: [email protected].

Motivation: Identification of novel therapeutic effects for existing US Food and Drug Administration (FDA)-approved drugs, drug repurposing, is an approach aimed to dramatically shorten the drug discovery process, which is costly, slow and risky. Several computational approaches use transcriptional data to find potential repurposing candidates. The main hypothesis of such approaches is that if gene expression signature of a particular drug is opposite to the gene expression signature of a disease, that drug may have a potential therapeutic effect on the disease. However, this may not be optimal since it fails to consider the different roles of genes and their dependencies at the system level. Results: We propose a systems biology approach to discover novel therapeutic roles for established drugs that addresses some of the issues in the current approaches. To do so, we use publicly available drug and disease data to build a drug-disease network by considering all interactions between drug targets and disease-related genes in the context of all known signaling pathways. This network is integrated with gene-expression measurements to identify drugs with new desired therapeutic effects based on a system-level analysis method. We compare the proposed approach with the drug repurposing approach proposed by Sirota et al. on four human diseases: idiopathic pulmonary fibrosis, non-small cell lung cancer, prostate cancer and breast cancer. We evaluate the proposed approach based on its ability to re-discover drugs that are already FDA-approved for a given disease. Availability and implementation: The R package DrugDiseaseNet is under review for publication in Bioconductor and is available at https://github.com/azampvd/DrugDiseaseNet. Supplementary information: Supplementary data are available at Bioinformatics online.

MOTIVATION: The ultimate goal of any experiment is to understand the biological phenomena underlying the condition investigated. This process often results in genes network through which a certain biological mechanism is explained. Such networks have been proven to be extremely useful, for the prediction of mechanisms of action of drugs or the responses of an organism to a specific impact (e.g. a disease, a treatment, etc.). Here, we introduce an approach able to build a network that captures the putative mechanisms at play in the given condition, by using datasets from multiple experiments studying the same phenotype. This method takes advantage of known interactions extracted from multiple sources such as protein-protein interactions and curated biological pathways. Based on such prior knowledge, we overcome the drawbacks of snap-shot data by considering the possible effects of each gene on its neighbors. RESULTS: We show the effectiveness of this approach in three different case studies and validate the results in two ways considering the identified genes and interactions between them. We compare our findings with the results of two widely-used methods in the same category as well as the classical approach of selecting differentially expressed (DE) genes in an investigated condition. The results show that 'neighbor-net' analysis is able to report biological mechanisms that are significantly relevant to the given diseases in all the three case studies, and performs better compared to all reference methods using both validation approaches. AVAILABILITY AND IMPLEMENTATION: The proposed method is implemented as in R and will be available an a Bioconductor package. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A crucial step in the understanding of any phenotype is the correct identification of the signaling pathways that are significantly impacted in that phenotype. However, most current pathway analysis methods produce both false positives as well as false negatives in certain circumstances. We hypothesized that such incorrect results are due to the fact that the existing methods fail to distinguish between the primary dis-regulation of a given gene itself and the effects of signaling coming from upstream. Furthermore, a modern whole-genome experiment performed with a next-generation technology spends a great deal of effort to measure the entire set of 30,000-100,000 transcripts in the genome. This is followed by the selection of a few hundreds differentially expressed genes, step that literally discards more than 99% of the collected data. We also hypothesized that such a drastic filtering could discard many genes that play crucial roles in the phenotype. We propose a novel topology-based pathway analysis method that identifies significantly impacted pathways using the entire set of measurements, thus allowing the full use of the data provided by NGS techniques. The results obtained on 24 real data sets involving 12 different human diseases, as well as on 8 yeast knock-out data sets show that the proposed method yields significant improvements with respect to the state-of-the-art methods: SPIA, GSEA and GSA. AVAILABILITY: Primary dis-regulation analysis is implemented in R and included in ROntoTools Bioconductor package (versions ≥ 2.0.0). https://www.bioconductor.org/packages/release/bioc/html/ROntoTools.html.

Identifying the pathways and mechanisms that are significantly impacted in a given phenotype is challenging. Issues include patient heterogeneity and noise. Many experiments do not have a large enough sample size to achieve the statistical power necessary to identify significantly impacted pathways. Meta-analysis based on combining p-values from individual experiments has been used to improve power. However, all classical meta-analysis approaches work under the assumption that the p-values produced by experiment-level statistical tests follow a uniform distribution under the null hypothesis. Here we show that this assumption does not hold for three mainstream pathway analysis methods, and significant bias is likely to affect many, if not all such meta-analysis studies. We introduce DANUBE, a novel and unbiased approach to combine statistics computed from individual studies. Our framework uses control samples to construct empirical null distributions, from which empirical p-values of individual studies are calculated and combined using either a Central Limit Theorem approach or the additive method. We assess the performance of DANUBE using four different pathway analysis methods. DANUBE is compared with five meta-analysis approaches, as well as with a pathway analysis approach that employs multiple datasets (MetaPath). The 25 approaches have been tested on 16 different datasets related to two human diseases, Alzheimer's disease (7 datasets) and acute myeloid leukemia (9 datasets). We demonstrate that DANUBE overcomes bias in order to consistently identify relevant pathways. We also show how the framework improves results in more general cases, compared to classical meta-analysis performed with common experiment-level statistical tests such as Wilcoxon and t-test.

MOTIVATION: The accumulation of high-throughput data in public repositories creates a pressing need for integrative analysis of multiple datasets from independent experiments. However, study heterogeneity, study bias, outliers and the lack of power of available methods present real challenge in integrating genomic data. One practical drawback of many P-value-based meta-analysis methods, including Fisher's, Stouffer's, minP and maxP, is that they are sensitive to outliers. Another drawback is that, because they perform just one statistical test for each individual experiment, they may not fully exploit the potentially large number of samples within each study. RESULTS: We propose a novel bi-level meta-analysis approach that employs the additive method and the Central Limit Theorem within each individual experiment and also across multiple experiments. We prove that the bi-level framework is robust against bias, less sensitive to outliers than other methods, and more sensitive to small changes in signal. For comparative analysis, we demonstrate that the intra-experiment analysis has more power than the equivalent statistical test performed on a single large experiment. For pathway analysis, we compare the proposed framework versus classical meta-analysis approaches (Fisher's, Stouffer's and the additive method) as well as against a dedicated pathway meta-analysis package (MetaPath), using 1252 samples from 21 datasets related to three human diseases, acute myeloid leukemia (9 datasets), type II diabetes (5 datasets) and Alzheimer's disease (7 datasets). Our framework outperforms its competitors to correctly identify pathways relevant to the phenotypes. The framework is sufficiently general to be applied to any type of statistical meta-analysis. AVAILABILITY AND IMPLEMENTATION: The R scripts are available on demand from the authors. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Oncogenes are known drivers of cancer phenotypes and targets of molecular therapies; however, the complex and diverse signaling mechanisms regulated by oncogenes and potential routes to targeted therapy resistance remain to be fully understood. To this end, we present an approach to infer regulatory mechanisms downstream of the HER2 driver oncogene in SUM-225 metastatic breast cancer cells from dynamic gene expression patterns using a succession of analytical techniques, including a novel MP grammars method to mathematically model putative regulatory interactions among sets of clustered genes. RESULTS: Our method highlighted regulatory interactions previously identified in the cell line and a novel finding that the HER2 oncogene, as opposed to the proto-oncogene, upregulates expression of the E2F2 transcription factor. By targeted gene knockdown we show the significance of this, demonstrating that cancer cell-matrix adhesion and outgrowth were markedly inhibited when E2F2 levels were reduced. Thus, validating in this context that upregulation of E2F2 represents a key intermediate event in a HER2 oncogene-directed gene expression-based signaling circuit. This work demonstrates how predictive modeling of longitudinal gene expression data combined with multiple systems-level analyses can be used to accurately predict downstream signaling pathways. Here, our integrated method was applied to reveal insights as to how the HER2 oncogene drives a specific cancer cell phenotype, but it is adaptable to investigate other oncogenes and model systems. AVAILABILITY AND IMPLEMENTATION: Accessibility of various tools is listed in methods; the Log-Gain Stoichiometric Stepwise algorithm is accessible at http://www.cbmc.it/software/Software.php.

The Great Lakes Bioinformatics Consortium (GLBC) is pleased to announce its ninth annual conference, the Great Lakes Bioinformatics Conference (GLBIO), to be held May 16–18, 2014 in Cincinnati, OH. GLBIO 2014 will be hosted by Cincinnati Children’s Hospital Medical Center in conjunction with the University of Cincinnati. The conference, an official conference of the International Society for Computational Biology (ISCB), provides an interdisciplinary forum for the discussion of research findings and methods, and development of long-term relationships and networking opportunities, for researchers within the region, as well as from around the world. The program will include oral presentations, poster presentations, invited keynote speakers and tutorials. From novice to expert, attendees partake in a variety of workshops, tutorials, presentations, posters, networking events and exhibits. Keynote speakers include Gary Bader, The Donnelly Centre at the University of Toronto; Tanya Y. Berger-Wolf, Department of Computer Science, University of Illinois; Charles Brooks, University of Michigan and Mike Hawrylycz, Allen Institute for Brain Science. Hands-on tutorials will be led by Sorin Draghici, Wayne State University (advanced gene network); Anil Jegga, Children’s Hospital Medical Center, University of Cincinnati (advanced gene network); Jarek Meller, Children’s Hospital Medical Center, University of Cincinnati (an introduction to bioinformatics); Isaac Neuhaus, University of CA at San Francisco (visualization web interface JavaScript for programmers); Larsson Omberg, Sage Bionetworks (an introduction to Sage Synapys platform) and more! Interested in being a part of this dynamic group of speakers? See the GLBIO 2014 Web site for details and links, at www.iscb.org/glbio. GLBIO 2014 will also feature professional and career development sessions sponsored by Federation of American Societies of Experimental Biology (FASEB) Minority Access to Research Careers (MARC). This unique program will include a career fair, wherein students will receive career guidance (including a curriculum vitae (CV) critique) and will have the opportunity to meet with recruiters. Conference educational session topics will include Algorithm Development and Machine Learning, Bioimage Analysis, Biological Networks, Chemical Biology, Disease Models and Molecular Medicine, Evolutionary, Comparative and Metagenomics and many more! GLBIO has established a strong reputation for building relationships among a nationally prominent bioscience research community, showcasing the North American Great Lakes region as a perfect place to conduct computer-aided research in the life sciences. Of the 250 attendees of GLBIO 2013, 70% of conference attendees said they would attend the 2014 conference and 96% stated they would recommend the GLBIO conference to a colleague. For more information, please go to www.iscb.org/glbio to learn more about this exciting and unique gathering of industry experts and collegiate giants. Mark your calendar today and join us in Cincinnati at GLBIO 2014! Interested in supporting GLBIO 2014 through sponsorship or exhibiting? Please contact Stacy Slagor, Director of Corporate Relations and Development, ISCB, at [email protected]. About the Great Lakes Bioinformatics Consortium: The Great Lakes Bioinformatics Consortium strives to enhance educational opportunities and research infrastructure throughout the region, to make the Great Lakes a world leader in bioinformatics and to facilitate new discoveries in data-intensive biological research. The annual research meeting (GLBIO) serves as an informal communication and networking forum for professional development. We believe that by bringing together the Great Lakes bioinformatics community on a regular basis, many new initiatives will be born. The GLBC foresees development of regional research center grant proposals to identify central strengths for research centers in the Great Lakes region and to create funded centers for bioinformatics research. Additionally, the GLBC envisions scholarship and training investments that are focused on developing talent within the Great Lakes region. About the International Society for Computational Biology : The ISCB (www.iscb.org) is the sole society representing computational biology and bioinformatics on a worldwide scale. ISCB serves a global community of >3000 scientists who are dedicated to advancing the scientific understanding of living systems through computation. It convenes the world’s experts and future leaders in top conferences, including the Intelligent Systems in Molecular Biology (ISMB) Conference, and features journals that promote discovery and expand access to computational biology and bioinformatics. It delivers valuable information about training, education, employment and other relevant news. ISCB also provides an influential voice on government policies and scientific policies that are important to its members and to the general public.

With the rapid advancements in our data acquisition capabilities and the increased availability of gene interaction databases a variety of pathway analysis tools have been proposed. However, all these methods are dependent on the quality of the available pathways. These pathways were designed to describe the general mechanism of a particular disease or biological process. The known pathways encompass the results of many biological experiments and even though they represent our current understanding of those particular biological processes, they are still generally considered sketchy and incomplete. One piece of information that is generally missing regards the role or importance of a gene in a given pathway which we refer to as the gene contribution. We propose here a method, based on genetic algorithms, to objectively quantify the contribution of each gene. Using a pool of 24 data sets from 12 different conditions divided in train and test groups, we show how an impact pathway analysis method achieves significantly better results with the newly estimated gene contributions when compared with both the initial default contributions, as well as randomly selected gene contributions.

MOTIVATION: Identification of every single genome present in a microbial sample is an important and challenging task with crucial applications. It is challenging because there are typically millions of cells in a microbial sample, the vast majority of which elude cultivation. The most accurate method to date is exhaustive single-cell sequencing using multiple displacement amplification, which is simply intractable for a large number of cells. However, there is hope for breaking this barrier, as the number of different cell types with distinct genome sequences is usually much smaller than the number of cells. RESULTS: Here, we present a novel divide and conquer method to sequence and de novo assemble all distinct genomes present in a microbial sample with a sequencing cost and computational complexity proportional to the number of genome types, rather than the number of cells. The method is implemented in a tool called Squeezambler. We evaluated Squeezambler on simulated data. The proposed divide and conquer method successfully reduces the cost of sequencing in comparison with the naïve exhaustive approach. AVAILABILITY: Squeezambler and datasets are available at http://compbio.cs.wayne.edu/software/squeezambler/.

Ensemble systems improve the generalization of single classifiers by aggregating the prediction of a set of base classifiers. Assessing classification reliability (posterior probability) is crucial in a number of applications, such as biomedical and diagnosis applications, where the cost of a misclassified input vector can be unacceptable high. Available methods are limited to either calibrate the posterior probability on an aggregated decision value or obtain a posterior probability for each base classifier and aggregate the result. We propose a method that takes advantage of the distribution of the decision values from the base classifiers to summarize a statistic which is subsequently used to generate the posterior probability. Three approaches are considered to fit the probabilistic output to the statistic: the standard Gaussian CDF, isotonic regression, and linear logistic. Even though this study focuses on a bagged support vector machine ensemble (Z‐bag), our approach is not limited by the aggregation method selected, the choice of base classifiers, nor the statistic used. Performance is assessed on one artificial and 12 real‐world data sets from the UCI Machine Learning Repository. Our approach achieves comparable or better generalization on accuracy and posterior estimation to existing ensemble calibration methods although lowering computational cost.

Multiple displacement amplification (MDA) is a fast non-PCR based isothermal DNA amplification method that can amplify small amounts of DNA samples to a reasonable quantity for genomic analysis. This technique is suitable for metagenomics and single cell genome sequencing and related analyses. The distribution of the coverage of the output amplicons for single cell MDA unlike the multiceli amplification is not uniform. This distribution is unknown, and the parameters that affect this amplification bias have not been studied thoroughly. To have a better understanding of the MDA reaction we have developed a simplified mathematical model and the corresponding simulation algorithm called MDAsim to obtain a generative model for the output amplicons. In this paper we will show that the output coverage of MDAsim matches the experimental coverage very well. Therefore, the combination of MDAsim and a sequencing simulator can be utilized for test and evaluation of single cell assemblers avoiding the burden of experimental sequencing. Our results suggest that modelling the MDA mechanism and simulation of such increasingly complex models may provide valuable insight into the MDA process, which in turn can be used to design more efficient MDA reactions.

Identification of the most impacted signaling pathways in a given condition is a crucial step in understanding the underlying biological mechanism. An impact analysis that is able to take in consideration the structure of a given signaling pathway was proposed to measure the impact on each pathway given a list of differentially expressed (DE) genes and their fold changes. Here, we investigated the utility of incorporating the individual gene significance in the impact analysis of signaling pathways. We propose two alternative models to incorporate the individual gene p-values and compare their performance over a pool of 24 datasets. In addition, the two new models offer the ability to work with the entire set of gene expression measurements, thus eliminating the need to select differentially expressed genes.

Many analysis techniques are currently available to identify the signaling pathways significantly impacted in a given condition. All these approaches calculate a p-value that aims to quantify the significance of the involvement of a given pathway in the condition under study. These p-values were thought to be related to the likelihood of their respective pathways being involved in the given condition, and to be independent. Here we show that this is not true, and that many pathways are not independent and that can considerably affect each other's p-values through a phenomenon we refer to as “cross-talk.” Thus, the significance of a given pathway in a given experiment has to be interpreted in the context of the other pathways that appear to be significant. Using real data, we show that in same cases pathways with significant classical p-values are not biologically meaningful, and that some biologically meaningful pathways with insignificant p-values become significant when the cross-talk effects of other pathways are removed. We show that this phenomenon is related to the amount of common genes between different pathways, affecting the most widely used methods for pathway analysis, and we propose an analysis technique that is able to correct the over-enrichment significance of a pathway when the cross-talk effects of other pathways are removed.

BACKGROUND: The identification of gene sets that are significantly impacted in a given condition based on microarray data is a crucial step in current life science research. Most gene set analysis methods treat genes equally, regardless how specific they are to a given gene set. RESULTS: In this work we propose a new gene set analysis method that computes a gene set score as the mean of absolute values of weighted moderated gene t-scores. The gene weights are designed to emphasize the genes appearing in few gene sets, versus genes that appear in many gene sets. We demonstrate the usefulness of the method when analyzing gene sets that correspond to the KEGG pathways, and hence we called our method Pathway Analysis with Down-weighting of Overlapping Genes (PADOG). Unlike most gene set analysis methods which are validated through the analysis of 2-3 data sets followed by a human interpretation of the results, the validation employed here uses 24 different data sets and a completely objective assessment scheme that makes minimal assumptions and eliminates the need for possibly biased human assessments of the analysis results. CONCLUSIONS: PADOG significantly improves gene set ranking and boosts sensitivity of analysis using information already available in the gene expression profiles and the collection of gene sets to be analyzed. The advantages of PADOG over other existing approaches are shown to be stable to changes in the database of gene sets to be analyzed. PADOG was implemented as an R package available at: http://bioinformaticsprb.med.wayne.edu/PADOG/or http://www.bioconductor.org.

High throughput technologies enable researchers to measure expression levels on a genomic scale. However, the correct and efficient biological interpretation of such voluminous data remains a challenging problem. Many tools have been developed for the analysis of GO terms that are over- or under-represented in a list of differentially expressed genes. However, a previously unexplored aspect is the identification of changes in the way various biological processes interact in a given condition with respect to a reference. Here, we present a novel approach that aims at identifying such interactions between biological processes that are significantly different in a given phenotype with respect to normal. The proposed technique uses vector-space representation, SVD-based dimensionality reduction, differential weighting, and bootstrapping to asses the significance of the interactions under the multiple and complex dependencies expected between the biological processes. We illustrate our approach on two real data sets involving breast and lung cancer. More than 88 percent of the interactions found by our approach were deemed to be correct by an extensive manual review of literature. An interesting subset of such interactions is discussed in detail and shown to have the potential to open new avenues for research in lung and breast cancer.

MOTIVATION: Many models and analysis of signaling pathways have been proposed. However, neither of them takes into account that a biological pathway is not a fixed system, but instead it depends on the organism, tissue and cell type as well as on physiological, pathological and experimental conditions. RESULTS: The Biological Connection Markup Language (BCML) is a format to describe, annotate and visualize pathways. BCML is able to store multiple information, permitting a selective view of the pathway as it exists and/or behave in specific organisms, tissues and cells. Furthermore, BCML can be automatically converted into data formats suitable for analysis and into a fully SBGN-compliant graphical representation, making it an important tool that can be used by both computational biologists and 'wet lab' scientists. AVAILABILITY AND IMPLEMENTATION: The XML schema and the BCML software suite are freely available under the LGPL for download at http://bcml.dc-atlas.net. They are implemented in Java and supported on MS Windows, Linux and OS X.

Two radically different approaches are currently available to identify the signaling pathways that are significantly impacted in a given condition: enrichment analysis and impact analysis. These approaches calculate a p-value that aims to quantify the significance of the involvement of the given pathway in the condition under study. These p-values were thought to be inversely proportional to the likelihood of their respective pathways being involved in the given condition, and hence be independent. Here we show that various pathways can affect each other's p-values in significant ways. Thus, the significance of a given pathway in a given experiment has to be interpreted in the context of the other pathways that appear to be significant. In certain circumstances, pathways previously found to be significant with some of the existing methods may not be so. We hypothesize that the phenomenon is related to the amount of common genes between different pathways. Here we present results obtained by analyzing pathways obtained from the KEGG signaling pathways database. However, the same phenomenon is expected to influence the analysis of any pathways from any pathway repository.

The correct interpretation of many molecular biology experiments depends in an essential way on the accuracy and consistency of the existing annotation databases. Such databases are meant to act as repositories for our biological knowledge as we acquire and refine it. Hence, by definition, they are incomplete at any given time. In this paper, we describe a technique that improves our previous method for predicting novel GO annotations by extracting implicit semantic relationships between genes and functions. In this work, we use a vector space model and a number of weighting schemes in addition to our previous latent semantic indexing approach. The technique described here is able to take into consideration the hierarchical structure of the Gene Ontology (GO) and can weight differently GO terms situated at different depths. The prediction abilities of 15 different weighting schemes are compared and evaluated. Nine such schemes were previously used in other problem domains, while six of them are introduced in this paper. The best weighting scheme was a novel scheme, n2tn. Out of the top 50 functional annotations predicted using this weighting scheme, we found support in the literature for 84 percent of them, while 6 percent of the predictions were contradicted by the existing literature. For the remaining 10 percent, we did not find any relevant publications to confirm or contradict the predictions. The n2tn weighting scheme also outperformed the simple binary scheme used in our previous approach.

MOTIVATION: Gene expression class comparison studies may identify hundreds or thousands of genes as differentially expressed (DE) between sample groups. Gaining biological insight from the result of such experiments can be approached, for instance, by identifying the signaling pathways impacted by the observed changes. Most of the existing pathway analysis methods focus on either the number of DE genes observed in a given pathway (enrichment analysis methods), or on the correlation between the pathway genes and the class of the samples (functional class scoring methods). Both approaches treat the pathways as simple sets of genes, disregarding the complex gene interactions that these pathways are built to describe. RESULTS: We describe a novel signaling pathway impact analysis (SPIA) that combines the evidence obtained from the classical enrichment analysis with a novel type of evidence, which measures the actual perturbation on a given pathway under a given condition. A bootstrap procedure is used to assess the significance of the observed total pathway perturbation. Using simulations we show that the evidence derived from perturbations is independent of the pathway enrichment evidence. This allows us to calculate a global pathway significance P-value, which combines the enrichment and perturbation P-values. We illustrate the capabilities of the novel method on four real datasets. The results obtained on these data show that SPIA has better specificity and more sensitivity than several widely used pathway analysis methods. AVAILABILITY: SPIA was implemented as an R package available at http://vortex.cs.wayne.edu/ontoexpress/

Like most classification techniques, the existing Support Vector Machines (SVM) approaches are challenged to correctly classify their input when the data points are either very close to the decision boundary or very dissimilar from the training data set. In both situations, most classifiers including SVMs will still give a prediction by assigning the test point to one of the classes. However, when a test instance is very close to the decision boundary, the side of the boundary on which the instance lies, and hence the predicted class, will depend in many instances more on the choices of the tuning or training parameters rather than a clear differences in features. Furthermore, if a test instance is substantially different from all instances used during the training, the classical SVM classifiers will still assign it to a class although there is little evidence to support such assignment. In both cases, it is very useful for a classifier to be able to assess its ability to classify a given instance by identifying those regions of the feature space in which the class assignments are less certain. In this paper, we propose two novel approaches based on: i) a geometric uncertainty margin and ii) the convex hulls of the training points in the feature space. Our proposed techniques improve upon the existing SVM-based approaches by adding the ability to identify “uncertainty” areas where the assignment of a test instance to a class cannot be guaranteed. We illustrate both the problems and our novel techniques on the Iris data set from the UCI machine learning repository.

MOTIVATION: The BioArray Software Environment (BASE) is a very popular MIAME-compliant, web-based microarray data repository. However in BASE, like in most other microarray data repositories, the experiment annotation and raw data uploading can be very timeconsuming, especially for large microarray experiments. RESULTS: We developed KUTE (Karmanos Universal daTabase for microarray Experiments), as a plug-in for BASE 2.0 that addresses these issues. KUTE provides an automatic experiment annotation feature and a completely redesigned data work-flow that dramatically reduce the human-computer interaction time. For instance, in BASE 2.0 a typical Affymetrix experiment involving 100 arrays required 4 h 30 min of user interaction time forexperiment annotation, and 45 min for data upload/download. In contrast, for the same experiment, KUTE required only 28 min of user interaction time for experiment annotation, and 3.3 min for data upload/download. AVAILABILITY: http://vortex.cs.wayne.edu/kute/index.html.

The correct interpretation of many molecular biology experiments depends in an essential way on the accuracy and consistency of the existing annotation databases. Such databases are meant to act as repositories for our biological knowledge as we acquire and refine it. Hence, by definition they are incomplete at any given time. In this paper we describe a technique that improves our previous method for extracting implicit semantic relationships between genes and functions. We added a number of weighting schemes to our previous latent semantic indexing approach. We used this technique to analyze the current annotations of the human genome. The predictions of 15 different weighting schemes were compared and evaluated. Out of the top 50 functional annotations predicted using the best performing weighting scheme, we found support in the literature for 82% of them. For 10% of our prediction we did not find any relevant publications, and 6% were actually contradicted by existing literature. This weighting scheme also outperformed the simple binary scheme used in our previous approach. Our method is independent of the organism and can be used to analyze and improve the quality of the data of any public or private annotation database

The term machine learning refers to a set of topics dealing with the creation and evaluation of algorithms that facilitate pattern recognition, classification, and prediction, based on models derived from existing data. Two facets of mechanization should be acknowledged when considering machine learning in broad terms. Firstly, it is intended that the classification and prediction tasks can be accomplished by a suitably programmed computing machine. That is, the product of machine learning is a classifier that can be feasibly used on available hardware. Secondly, it is intended that the creation of the classifier should itself be highly mechanized, and should not involve too much human input. This second facet is inevitably vague, but the basic objective is that the use of automatic algorithm construction methods can minimize the possibility that human biases could affect the selection and performance of the algorithm. Both the creation of the algorithm and its operation to classify objects or predict events are to be based on concrete, observable data. The history of relations between biology and the field of machine learning is long and complex. An early technique [1] for machine learning called the perceptron constituted an attempt to model actual neuronal behavior, and the field of artificial neural network (ANN) design emerged from this attempt. Early work on the analysis of translation initiation sequences [2] employed the perceptron to define criteria for start sites in Escherichia coli. Further artificial neural network architectures such as the adaptive resonance theory (ART) [3] and neocognitron [4] were inspired from the organization of the visual nervous system. In the intervening years, the flexibility of machine learning techniques has grown along with mathematical frameworks for measuring their reliability, and it is natural to hope that machine learning methods will improve the efficiency of discovery and understanding in the mounting volume and complexity of biological data. This tutorial is structured in four main components. Firstly, a brief section reviews definitions and mathematical prerequisites. Secondly, the field of supervised learning is described. Thirdly, methods of unsupervised learning are reviewed. Finally, a section reviews methods and examples as implemented in the open source data analysis and visualization language R (http://www.r-project.org).

MOTIVATION: Annotation databases are widely used as public repositories of biological knowledge. However, most of these resources have been developed by independent groups which used different designs and different identifiers for the same biological entities. As we show in this article, incoherent name spaces between various databases represent a serious impediment to using the existing annotations at their full potential. Navigating between various such name spaces by mapping IDs from one database to another is a very important issue which is not properly addressed at the moment. RESULTS: We have developed a web-based resource, Onto-Translate (OT), which effectively addresses this problem. OT is able to map onto each other different types of biological entities from the following annotation databases: Swiss-Prot, TrEMBL, NREF, PIR, Gene Ontology, KEGG, Entrez Gene, GenBank, GenPept, IMAGE, RefSeq, UniGene, OMIM, PDB, Eukaryotic Promoter Database, HUGO Gene Nomenclature Committee and NetAffx. Currently, OT is able to perform 462 types of mappings between 29 different types of IDs from 17 databases concerning 53 organisms. Among these, over 300 types of translations and 15 types of IDs are not currently supported by any other tool or resource. On average, OT is able to correctly map between 96 and 99% of the biological entities provided as input. In terms of speed, sets of approximately 20 000 IDs can be translated in <30 s, in most cases. AVAILABILITY: OT is a part of Onto-Tools, which is freely available at http://vortex.cs.wayne.edu/Projects.html

Independent of the platform and the analysis methods used, the result of a microarray experiment is, in most cases, a list of differentially expressed genes. An automatic ontological analysis approach has been recently proposed to help with the biological interpretation of such results. Currently, this approach is the de facto standard for the secondary analysis of high throughput experiments and a large number of tools have been developed for this purpose. We present a detailed comparison of 14 such tools using the following criteria: scope of the analysis, visualization capabilities, statistical model(s) used, correction for multiple comparisons, reference microarrays available, installation issues and sources of annotation data. This detailed analysis of the capabilities of these tools will help researchers choose the most appropriate tool for a given type of analysis. More importantly, in spite of the fact that this type of analysis has been generally adopted, this approach has several important intrinsic drawbacks. These drawbacks are associated with all tools discussed and represent conceptual limitations of the current state-of-the-art in ontological analysis. We propose these as challenges for the next generation of secondary data analysis tools.

The correct interpretation of any biological experiment depends in an essential way on the accuracy and consistency of the existing annotation databases. Such databases are ubiquitous and used by all life scientists in most experiments. However, it is well known that such databases are incomplete and many annotations may also be incorrect. In this paper we describe a technique that can be used to analyze the semantic content of such annotation databases. Our approach is able to extract implicit semantic relationships between genes and functions. This ability allows us to discover novel functions for known genes. This approach is able to identify missing and inaccurate annotations in existing annotation databases, and thus help improve their accuracy. We used our technique to analyze the current annotations of the human genome. From this body of annotations, we were able to predict 212 additional gene-function assignments. A subsequent literature search found that 138 of these gene-functions assignments are supported by existing peer-reviewed papers. An additional 23 assignments have been confirmed in the meantime by the addition of the respective annotations in later releases of the Gene Ontology database. Overall, the 161 confirmed assignments represent 75.95% of the proposed gene-function assignments. Only one of our predictions (0.4%) was contradicted by the existing literature. We could not find any relevant articles for 50 of our predictions (23.58%). The method is independent of the organism and can be used to analyze and improve the quality of the data of any public or private annotation database.

BACKGROUND: cDNA microarrays are a powerful means to screen for biologically relevant gene expression changes, but are often limited by their ability to detect small changes accurately due to "noise" from random and systematic errors. While experimental designs and statistical analysis methods have been proposed to reduce these errors, few studies have tested their accuracy and ability to identify small, but biologically important, changes. Here, we have compared two cDNA microarray experimental design methods with northern blot confirmation to reveal changes in gene expression that could contribute to the early antiproliferative effects of neuregulin on MCF10AT human breast epithelial cells. RESULTS: We performed parallel experiments on identical samples using a dye-swap design with ANOVA and an experimental design that excludes systematic biases by "correcting" experimental/control hybridization ratios with control/control hybridizations on a spot-by-spot basis. We refer to this approach as the "control correction method" (CCM). Using replicate arrays, we identified a decrease in proliferation genes and an increase in differentiation genes. Using an arbitrary cut-off of 1.7-fold and p values <0.05, we identified a total of 32 differentially expressed genes, 9 with the dye-swap method, 18 with the CCM, and 5 genes with both methods. 23 of these 32 genes were subsequently verified by northern blotting. Most of these were <2-fold changes. While the dye-swap method (using either ANOVA or Bayesian analysis) detected a smaller number of genes (14-16) compared to the CCM (46), it was more accurate (89-92% vs. 75%). Compared to the northern blot results, for most genes, the microarray results underestimated the fold change, implicating the importance of detecting these small changes. CONCLUSIONS: We validated two experimental design paradigms for cDNA microarray experiments capable of detecting small (<2-fold) changes in gene expression with excellent fidelity that revealed potentially important genes associated with the anti-proliferative effects of neuregulin on MCF10AT breast epithelial cells.

MOTIVATION: We implement a data mining technique based on the method of Independent Component Analysis (ICA) to generate reliable independent data sets for different HIV therapies. We show that this technique takes advantage of the ICA power to eliminate the noise generated by artificial interaction of HIV system dynamics. Moreover, the incorporation of the actual laboratory data sets into the analysis phase offers a powerful advantage when compared with other mathematical procedures that consider the general behavior of HIV dynamics. RESULTS: The ICA algorithm has been used to generate different patterns of the HIV dynamics under different therapy conditions. The Kohonen Map has been used to eliminate redundant noise in each pattern to produce a reliable data set for the simulation phase. We show that under potent antiretroviral drugs, the value of the CD4+ cells in infected persons decreases gradually by about 11% every 100 days and the levels of the CD8+ cells increase gradually by about 2% every 100 days. AVAILABILITY: Executable code and data libraries are available by contacting the corresponding author. IMPLEMENTATION: Mathematica 4 has been used to simulate the suggested model. A Pentium III or higher platform is recommended.

MOTIVATION: A crucial step in microarray data analysis is the selection of subsets of interesting genes from the initial set of genes. In many cases, especially when comparing a specific condition to a reference, the genes of interest are those which are differentially expressed. Two common methods for gene selection are: (a) selection by fold difference (at least n fold variation) and (b) selection by altered ratio (at least n standard deviations away from the mean ratio). RESULTS: The novel method proposed here is based on ANOVA and uses replicate spots to estimate an empirical distribution of the noise. The measured intensity range is divided in a number of intervals. A noise distribution is constructed for each such interval. Bootstrapping is used to map the desired confidence levels from the noise distribution corresponding to a given interval to the measured log ratios in that interval. If the method is applied on individual arrays having replicate spots, the method can calculate an overall width of the noise distribution which can be used as an indicator of the array quality. We compared this method with the fold change and unusual ratio method. We also discuss the relationship with an ANOVA model proposed by Churchill et al. In silico experiments were performed while controlling the degree of regulation as well as the amount of noise. Such experiments show the performance of the classical methods can be very unsatisfactory. We also compared the results of the 2-fold method with the results of the noise sampling method using pre and post immortalization cell lines derived from the MDAH041 fibroblasts hybridized on Affymetrix GeneChip arrays. The 2-fold method reported 198 genes as upregulated and 493 genes as downregulated. The noise sampling method reported 98 gene upregulated and 240 genes downregulated at the 99.99% confidence level. The methods agreed on 221 genes downregulated and 66 genes upregulated. Fourteen genes from the subset of genes reported by both methods were all confirmed by Q-RT-PCR. Alternative assays on various subsets of genes on which the two methods disagreed suggested that the noise sampling method is likely to provide fewer false positives.

MOTIVATION: Drug resistance is a very important factor influencing the failure of current HIV therapies. The ability to predict the drug resistance of HIV protease mutants may be useful in developing more effective and longer lasting treatment regimens. METHODS: The HIV resistance is predicted to two current protease inhibitors, Indinavir and Saquinavir. The problem was approached from two perspectives. First, a predictor was constructed based on the structural features of the HIV protease-drug inhibitor complex. A particular structure was represented by its list of contacts between the inhibitor and the protease. Next, a classifier was constructed based on the sequence data of various drug resistant mutants. In both cases, self-organizing maps were first used to extract the important features and cluster the patterns in an unsupervised manner. This was followed by subsequent labelling based on the known patterns in the training set. RESULTS: The prediction performance of the classifiers was measured by cross-validation. The classifier using the structure information correctly classified previously unseen mutants with an accuracy of between 60 and 70%. Several architectures were tested on the more abundant sequence data. The best single classifier provided an accuracy of 68% and a coverage of 69%. Multiple networks were then combined into various majority voting schemes. The best combination yielded an average of 85% coverage and 78% accuracy on previously unseen data. This is more than two times better than the 33% accuracy expected from a random classifier.

This paper presents a brief review of some analog hardware implementations of neural networks. Several criteria for the classification of general neural networks implementations are discussed and a taxonomy induced by these criteria is presented. The paper also discusses some characteristics of analog implementations as well as some trade-offs and issues identified in the work reviewed. Parameters such as precision, chip area, power consumption, speed and noise susceptibility are discussed in the context of neural implementations. A unified review of various "VLSI friendly" algorithms is also presented. The paper concludes with some conclusions drawn from the analysis of the implementations presented.

This paper analyzes some aspects of the computational power of neural networks (NN) using integer weights in a very restricted range. Using limited range integer values opens the road for efficient VLSI implementations because: 1) a limited range for the weights can be translated into reduced storage requirements, and 2) integer computation can be implemented in a more efficient way than the floating point one. The paper shows that a neural network using integer weights in the range [-p,p] (where p is a small integer value) can classify correctly any set of patterns included in a hypercube of unit side length centered around the origin of R/sub n/, n/spl ges/2, for which the minimum Euclidean distance between two patterns of opposite classes is d/sub min//spl ges//spl radic/(n-1)/2p.

The VLSI-friendly constraint based decomposition (VCBD) algorithm guarantees construction of neural network classifiers with a user-chosen maximum neuron fan-in and weight values confined to a narrow band of integers. Networks with these characteristics can be efficiently implemented in VLSI. This paper presents the design of a CMOS neuron that fully exploits the characteristics of solutions produced by VCBD. The limited integer weight range enables efficient use of current-mode techniques for addition and multiplication operations within the neuron. Weights are stored in digital form without quantization error. A simple current comparator is used to provide the neuron output. Circuit operation has been verified using SPICE and these simulation results are included.

This paper presents a neural network based artificial vision system able to analyze the image of a car given by a camera, locate the registration plate and recognize the registration number of the car. The paper describes in detail various practical problems encountered in implementing this particular application and the solutions used to solve them. The main features of the system presented are: controlled stability-plasticity behavior, controlled reliability threshold, both off-line and on-line learning, self assessment of the output reliability and high reliability based on high level multiple feedback. The system has been designed using a modular approach. Sub-modules can be upgraded and/or substituted independently, thus making the system potentially suitable in a large variety of vision applications. The OCR engine was designed as an interchangeable plug-in module. This allows the user to choose an OCR engine which is suited to the particular application and to upgrade it easily in the future. At present, there are several versions of this OCR engine. One of them is based on a fully connected feedforward artificial neural network with sigmoidal activation functions. This network can be trained with various training algorithms such as error backpropagation. An alternative OCR engine is based on the constraint based decomposition (CBD) training architecture. The system has showed the following performances (on average) on real-world data: successful plate location and segmentation about 99%, successful character recognition about 98% and successful recognition of complete registration plates about 80%.