Cancer is a leading cause of death in the US, second only to heart disease. It is primarily a result of a combination of an estimated two-nine genetic mutations (multi-hit combinations). Although a body of research has identified hundreds of cancer-causing genetic mutations, we don't know the specific combination of mutations responsible for specific instances of cancer for most cancer types. An approximate algorithm for solving the weighted set cover problem was previously adapted to identify combinations of genes with mutations that may be responsible for individual instances of cancer. However, the algorithm's computational requirement scales exponentially with the number of genes, making it impractical for identifying more than three-hit combinations, even after the algorithm was parallelized and scaled up to a V100 GPU. Since most cancers have been estimated to require more than three hits, we scaled out the algorithm to identify combinations of four or more hits using 1000 nodes (6000 V100 GPUs with ≈ 48×106processing cores) on the Summit supercomputer at Oak Ridge National Laboratory. Efficiently scaling out the algorithm required a series of algorithmic innovations and optimizations for balancing an exponentially divergent workload across processors and for minimizing memory latency and inter-node communication. We achieved an average strong scaling efficiency of 90.14% (80.96%-97.96% for 200 to 1000 nodes), compared to a 100 node run, with 84.18% scaling efficiency for 1000 nodes. With experimental validation, the multi-hit combinations identified here could provide further insight into the etiology of different cancer subtypes and provide a rational basis for targeted combination therapy.

Individual instances of cancer are primarily a result of a combination of a small number of genetic mutations (hits). Knowing the number of such mutations is a prerequisite for identifying specific combinations of carcinogenic mutations and understanding the etiology of cancer. We present a mathematical model for estimating the number of hits based on the distribution of somatic mutations. The model is fundamentally different from previous approaches, which are based on cancer incidence by age. Our somatic mutation based model is likely to be more robust than age-based models since it does not require knowing or accounting for the highly variable mutation rate, which can vary by over three orders of magnitude. In fact, we find that the number of somatic mutations at diagnosis is weakly correlated with age at cancer diagnosis, most likely due to the extreme variability in mutation rates between individuals. Comparing the distribution of somatic mutations predicted by our model to the actual distribution from 6904 tumor samples we estimate the number of hits required for carcinogenesis for 17 cancer types. We find that different cancer types exhibit distinct somatic mutational profiles corresponding to different numbers of hits. Why might different cancer types require different numbers of hits for carcinogenesis? The answer may provide insight into the unique etiology of different cancer types.

MOTIVATION: Inferring lengths of inherited microsatellite alleles with single base pair resolution from short sequence reads is challenging due to several sources of noise caused by the repetitive nature of microsatellites and the technologies used to generate raw sequence data. RESULTS: We have developed a program, GenoTan, using a discretized Gaussian mixture model combined with a rules-based approach to identify inherited variation of microsatellite loci from short sequence reads without paired-end information. It effectively distinguishes length variants from noise including insertion/deletion errors in homopolymer runs by addressing the bidirectional aspect of insertion and deletion errors in sequence reads. Here we first introduce a homopolymer decomposition method which estimates error bias toward insertion or deletion in homopolymer sequence runs. Combining these approaches, GenoTan was able to genotype 94.9% of microsatellite loci accurately from simulated data with 40x sequence coverage quickly while the other programs showed <90% correct calls for the same data and required 5∼30× more computational time than GenoTan. It also showed the highest true-positive rate for real data using mixed sequence data of two Drosophila inbred lines, which was a novel validation approach for genotyping. AVAILABILITY: GenoTan is open-source software available at http://genotan.sourceforge.net.

MOTIVATION: Simple tandem repeats are highly variable genetic elements and widespread in genomes of many organisms. Next-generation sequencing technologies have enabled a robust comparison of large numbers of simple tandem repeat loci; however, analysis of their variation using traditional sequence analysis approaches still remains limiting and problematic due to variants occurring in repeat sequences confusing alignment programs into mapping sequence reads to incorrect loci when the sequence reads are significantly different from the reference sequence. RESULTS: We have developed a program, ReviSTER, which is an automated pipeline using a 'local mapping reference reconstruction method' to revise mismapped or partially misaligned reads at simple tandem repeat loci. RevisSTER estimates alleles of repeat loci using a local alignment method and creates temporary local mapping reference sequences, and finally remaps reads to the local mapping references. Using this approach, ReviSTER was able to successfully revise reads misaligned to repeat loci from both simulated data and real data. AVAILABILITY: ReviSTER is open-source software available at http://revister.sourceforge.net. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Document similarity metrics such as PubMed's 'Find related articles' feature, which have been primarily used to identify studies with similar topics, can now also be used to detect duplicated or potentially plagiarized papers within literature reference databases. However, the CPU-intensive nature of document comparison has limited MEDLINE text similarity studies to the comparison of abstracts, which constitute only a small fraction of a publication's total text. Extending searches to include text archived by online search engines would drastically increase comparison ability. For large-scale studies, submitting short phrases encased in direct quotes to search engines for exact matches would be optimal for both individual queries and programmatic interfaces. We have derived a method of analyzing statistically improbable phrases (SIPs) for assistance in identifying duplicate content. RESULTS: When applied to MEDLINE citations, this method substantially improves upon previous algorithms in the detection of duplication citations, yielding a precision and recall of 78.9% (versus 50.3% for eTBLAST) and 99.6% (versus 99.8% for eTBLAST), respectively. AVAILABILITY: Similar citations identified by this work are freely accessible in the Déjà vu database, under the SIP discovery method category at http://dejavu.vbi.vt.edu/dejavu/.

BACKGROUND: Mass spectrometry-based biomarker discovery has long been hampered by the difficulty in reconciling lists of discriminatory peaks identified by different laboratories for the same diseases studied. We describe a multi-statistical analysis procedure that combines several independent computational methods. This approach capitalizes on the strengths of each to analyze the same high-resolution mass spectral data set to discover consensus differential mass peaks that should be robust biomarkers for distinguishing between disease states. RESULTS: The proposed methodology was applied to a pilot narcolepsy study using logistic regression, hierarchical clustering, t-test, and CART. Consensus, differential mass peaks with high predictive power were identified across three of the four statistical platforms. Based on the diagnostic accuracy measures investigated, the performance of the consensus-peak model was a compromise between logistic regression and CART, which produced better models than hierarchical clustering and t-test. However, consensus peaks confer a higher level of confidence in their ability to distinguish between disease states since they do not represent peaks that are a result of biases to a particular statistical algorithm. Instead, they were selected as differential across differing data distribution assumptions, demonstrating their true discriminatory potential. CONCLUSION: The methodology described here is applicable to any high-resolution MALDI mass spectrometry-derived data set with minimal mass drift which is essential for peak-to-peak comparison studies. Four statistical approaches with differing data distribution assumptions were applied to the same raw data set to obtain consensus peaks that were found to be statistically differential between the two groups compared. These consensus peaks demonstrated high diagnostic accuracy when used to form a predictive model as evaluated by receiver operating characteristics curve analysis. They should demonstrate a higher discriminatory ability as they are not biased to a particular algorithm. Thus, they are prime candidates for downstream identification and validation efforts.

MOTIVATION: Duplicate publication impacts the quality of the scientific corpus, has been difficult to detect, and studies this far have been limited in scope and size. Using text similarity searches, we were able to identify signatures of duplicate citations among a body of abstracts. RESULTS: A sample of 62,213 Medline citations was examined and a database of manually verified duplicate citations was created to study author publication behavior. We found that 0.04% of the citations with no shared authors were highly similar and are thus potential cases of plagiarism. 1.35% with shared authors were sufficiently similar to be considered a duplicate. Extrapolating, this would correspond to 3500 and 117,500 duplicate citations in total, respectively. AVAILABILITY: eTBLAST, an automated citation matching tool, and Déjà vu, the duplicate citation database, are freely available at http://invention.swmed.edu/ and http://spore.swmed.edu/dejavu

BACKGROUND: Single Nucleotide Polymorphisms (SNPs) are the most abundant form of genomic variation and can cause phenotypic differences between individuals, including diseases. Bases are subject to various levels of selection pressure, reflected in their inter-species conservation. RESULTS: We propose a method that is not dependant on transcription information to score each coding base in the human genome reflecting the disease probability associated with its mutation. Twelve factors likely to be associated with disease alleles were chosen as the input for a support vector machine prediction algorithm. The analysis yielded 83% sensitivity and 84% specificity in segregating disease like alleles as found in the Human Gene Mutation Database from non-disease like alleles as found in the Database of Single Nucleotide Polymorphisms. This algorithm was subsequently applied to each base within all known human genes, exhaustively confirming that interspecies conservation is the strongest factor for disease association. For each gene, the length normalized average disease potential score was calculated. Out of the 30 genes with the highest scores, 21 are directly associated with a disease. In contrast, out of the 30 genes with the lowest scores, only one is associated with a disease as found in published literature. The results strongly suggest that the highest scoring genes are enriched for those that might contribute to disease, if mutated. CONCLUSION: This method provides valuable information to researchers to identify sensitive positions in genes that have a high disease probability, enabling them to optimize experimental designs and interpret data emerging from genetic and epidemiological studies.

MOTIVATION: Microarrays rapidly generate large quantities of gene expression information, but interpreting such data within a biological context is still relatively complex and laborious. New methods that can identify functionally related genes via shared literature concepts will be useful in addressing these needs. RESULTS: We have developed a novel method that uses implicit literature relationships (concepts related via shared, intermediate concepts) to cluster related genes. Genes are evaluated for implicit connections within a network of biomedical objects (other genes, ontological concepts and diseases) that are connected via their co-occurrences in Medline titles and/or abstracts. On the basis of these implicit relationships, individual gene pairs are scored using a probability-based algorithm. Scores are generated for all pairwise combinations of genes, which are then clustered based on the scores. We applied this method to a test set composed of nine functional groups with known relationships. The method scored highly for all nine groups and significantly better than a benchmark co-occurrence-based method for six groups. We then applied this method to gene sets specific to two previously defined breast tumor subtypes. Analysis of the results recapitulated known biological relationships and identified novel pathway relationships unique to each tumor subtype. We demonstrate that this method provides a valuable new means of identifying and visualizing significantly related genes within gene lists via their implicit relationships in the literature.

BACKGROUND: RNA interference has revolutionized our ability to study the effects of altering the expression of single genes in mammalian (and other) cells through targeted knockdown of gene expression. In this report we describe a web-based computational tool, siRNA Information Resource (sIR), which consists of a new open source database that contains validation information about published siRNA sequences and also provides a user-friendly interface to design and analyze siRNA sequences against a chosen target sequence. RESULTS: The siRNA design tool described in this paper employs empirically determined rules derived from a meta-analysis of the published data; it uses a weighted scoring system that determines the optimal sequence within a target mRNA and thus aids in the rational selection of siRNA sequences. This scoring system shows a non-linear correlation with the knockdown efficiency of siRNAs. sIR provides a fast, customized BLAST output for all selected siRNA sequences against a variety of databases so that the user can verify the uniqueness of the design. We have pre-designed siRNAs for all the known human genes (24,502) in the Refseq database. These siRNAs were pre-BLASTed against the human Unigene database to estimate the target specificity and all results are available online. CONCLUSION: Although most of the rules for this scoring system were influenced by previously published rules, the weighted scoring system provides better flexibility in designing an appropriate siRNA when compared to the un-weighted scoring system. sIR is not only a comprehensive tool used to design siRNA sequences and lookup pre-designed siRNAs, but it is also a platform where researchers can share information on siRNA design and use.

MOTIVATION: The most widely used literature search techniques, such as those offered by NCBI's PubMed system, require significant effort on the part of the searcher, and inexperienced searchers do not use these systems as effectively as experienced users. Improved literature search engines can save researchers time and effort by making it easier to locate the most important and relevant literature. RESULTS: We have created and optimized a new, hybrid search system for Medline that takes natural text as input and then delivers results with high precision and recall. The combination of a fast, low-sensitivity weighted keyword-based first pass algorithm to cast a wide net to gather an initial set of literature, followed by a unique sentence-alignment based similarity algorithm to rank order those results was developed that is sensitive, fast and easy to use. Several text similarity search algorithms, both standard and novel, were implemented and tested in order to determine which obtained the best results in information retrieval exercises. AVAILABILITY: Literature searching algorithms are implemented in a system called eTBLAST, freely accessible over the web at http://invention.swmed.edu. A variety of other derivative systems and visualization tools provides the user with an enhanced experience and additional capabilities. CONTACT: [email protected].

MOTIVATION: While the mechanism for regulating alternative splicing is poorly understood, secondary structure has been shown to be integral to this process. Due to their propensity for forming complementary hairpin loops and their elevated mutation rates, tandem repeated sequences have the potential to influence splicing regulation. RESULTS: An analysis of human intronic sequences reveals a strong correlation between alternative splicing and the prevalence of mono- through hexanucleotide tandem repeats that may engage in complementary pairing in introns that flank alternatively spliced exons. While only 44% of the 18 173 genes in the Human Alternative Splicing Database are known to be alternatively spliced, they contain 84% of the 694 237 intronic complementary repeat pairs. Significantly, the normalized frequency and distribution of repeat sequences, independent of their potential for pairing, are indistinguishable between alternatively spliced and non-alternatively spliced genes. Thus, the increased prevalence of repeats with pairing potential in alternatively spliced genes is not merely a consequence of more repeats or repeat composition bias. These results suggest that complementary repeats may play a role in the regulation of alternative splicing. CONTACT: [email protected].

MOTIVATION: New relationships are often implicit from existing information, but the amount and growth of published literature limits the scope of analysis an individual can accomplish. Our goal was to develop and test a computational method to identify relationships within scientific reports, such that large sets of relationships between unrelated items could be sought out and statistically ranked for their potential relevance as a set. RESULTS: We first construct a network of tentative relationships between 'objects' of biomedical research interest (e.g. genes, diseases, phenotypes, chemicals) by identifying their co-occurrences within all electronically available MEDLINE records. Relationships shared by two unrelated objects are then ranked against a random network model to estimate the statistical significance of any given grouping. When compared against known relationships, we find that this ranking correlates with both the probability and frequency of object co-occurrence, demonstrating the method is well suited to discover novel relationships based upon existing shared relationships. To test this, we identified compounds whose shared relationships predicted they might affect the development and/or progression of cardiac hypertrophy. When laboratory tests were performed in a rodent model, chlorpromazine was found to reduce the progression of cardiac hypertrophy.

MOTIVATION: There is a general scientific need to be able to identify and evaluate what any given set of 'objects' (e.g. genes, phenotypes, chemicals, diseases) has in common. Whether it is to classify, expand upon or identify commonalities and functional groupings, informational needs can be diverse and the best source to identify relationships among a potentially heterogeneous set of objects is the scientific literature. RESULTS: We first establish a network of related objects by their co-occurrence within MEDLINE records. A set of objects within this network can then be queried to identify shared relationships, and a method is presented to score their statistical relevance by comparing observed frequencies with what would be expected in a random network model. Using Gene Ontology (GO) categories, we demonstrate that this method enables a quantitative ranking of the 'cohesiveness' of a set of objects and, importantly, allows other objects related to this set to be identified and evaluated for their 'cohesion' to it. Supplemental information: A list of ranked genes related to each GO category analyzed can be found at http://innovation.swmed.edu/IRIDESCENT/GO_relationships.htm

With the completion of the first genome, the daily analysis of genomic variation and the transcriptome, and new technologies amassing proteome data, the need for computational tools, inspired by biomedical insight and need are being developed by many groups to help everybody reduce this data to knowledge. To attain understanding from this data, several things are now required for progress; applied computational tools, access to phenotype presenting patients/biological material (and their genotype), and experts to assimilate this information. In this talk, I will describe efforts within our laboratory to develop and then validate some applied computational tools that are then used by our expert collaborators and us. These include: 1) Computer algorithms and codes that inspect the genome (coding and intronic) for highly probable simple sequence repeat and single nucleotide polymorphisms; 2) Techniques and applications to better identify information in text databases; 3) Software to integrate data from some of the many valuable public databases with experimental data (expression and polymorphism), and 4) Software to do integrated analysis and interactive visualization of genomic sequence dat.Each of our software applications is applied to biomedical problems to validate their utility and verify their accuracy, which can only be done by returning to the wet laboratory. The biomedical areas under investigation include cancer, cardiac disease, development, inflammation and infection. This approach has led to gene discoveries, phenotype causing polymorphisms and has inspired research directions for which the final answer is not yet in.Bioinformatics tools often can generate many more leads than are possible to pursue with existing technology. This has required us to develop some new technologies and laboratory techniques. For example, we have developed a method and apparatus that is in daily use for the synthesis of custom, software-directed high-density oligonucleotide arrays used for re-sequencing, methlyation analysis and comparative genomic hybridization. This device, called DOC (Digital Optical Chemistry), enables us to manufacture Affymetrix-style arrays with up to 192,000 features. Other apparatus, such as high-throughput oligo synthesizers, hyperspectral imaging scanners/microscopes and array spotting robots were also created as needed.This presentation will focus on our applied computational tools and the biomedical observations made with them and touch on how some of the hardware technologies have accelerated that process.Much of the software and databases we develop are available for public use via our www site at http://innovation.swmed.edu/. This work is supported by the NIH, the State of Texas and the P.O'B. Montgomery Distinguished Chair.

ARROGANT (ARRay OrGANizing Tool) is a software tool developed to facilitate the identification, annotation and comparison of large collections of genes or clones. The objective is to enable users to compile gene/clone collections from different databases, allowing them to design experiments and analyze the collections as well as associated experimental data efficiently. ARROGANT can relate different sequence identifiers to their common reference sequence using the UniGene database, allowing for the comparison of data from two different microarray experiments. ARROGANT has been successfully used to analyze microarray expression data for colon cancer, to compile genes potentially related to cardiac diseases for subsequent resequencing (to identify single nucleotide polymorphisms, SNPs), to design a new comprehensive human cDNA microarray for cancer, to combine and compare expression data generated by different microarrays and to provide annotation for genes on custom and Affymetrix chips.

SUMMARY: Microarray data management and processing (MAD) is a set of Windows integrated software for microarray analysis. It consists of a relational database for data storage with many user-interfaces for data manipulation, several text file parsers and Microsoft Excel macros for automation of data processing, and a generator to produce text files that are ready for cluster analysis. AVAILABILITY: Executable is available free of charge on http://pompous.swmed.edu. The source code is also available upon request.