Bioinformatics Choices Substantially Impact Isoform Analysis of RNA-seq Data from a Toxicogenomics Study
Joshua Xu (1), Xi Chen (2), Suresh Subraman (1), Binsheng Gong(1) and Weida Tong (1)

(1) Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, AR; (2) Department of Histology and Embryology, Harbin Medical University, Harbin, Heilongjiang, China

Alternative splicing events greatly increase the diversity of proteins which might response differently to toxic insults. Identification and quantification of isoforms are important for toxicological research for an improved understanding of the underlying mechanisms of toxicity. The advent of RNA-seq technology and related bioinformatics tools enable detection of novel isoforms and quantification of relative transcript abundances and thus enhance the research of alternative splicing. However, recent works have shown highly inconsistent results for quantitative analysis and there is no clear understanding of the source of variation. In this study, we investigated the impact of five potential factors on isoform detection and quantification, including the choice of bioinformatics methods, sequencing depth, library preparation, transcript abundance and treatment effect. Liver RNA samples from six rats (three treated by aflatoxin B1 for 5 days and three matched controls) were profiled with two libraries prepared for each sample. The first batch of libraries was sequenced twice while the second batch sequenced only once. Five bioinformatics pipelines were used, with two mapping tools (TopHat2 and STAR) followed by three isoform analysis approaches (Cufflinks, IsoLasso and FlipFlop). We evaluated the consistency of isoform detection for each potential factor and compared the differential analysis results of isoform expression and splicing for each factor. In summary, the choice of bioinformatics pipelines has a substantial impact on all aspects of quantitative isoform analysis of RNA-seq data, including the number of isoforms detected, differentially expressed isoforms, and their low concordance (about 41% on average) between pipelines. Differentially used transcription starting sites or alternative splicing events for the short-term aflatoxin B1 treatment cannot be reliably detected.
AR-BIC-4

Potential Reuse of Oncologic Drugs for the Treatment of Rare Diseases
Zhichao Liu, Hong Fang, William Slikker and Weida Tong

Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, AR

Cancer research has been a focus in the biomedical field resulting in many oncologic drugs in clinical use. In contrast, very few treatment options are available for rare diseases although they are progressive, disabling and life threatening. Therefore, we investigated the potential use of oncologic drugs for the treatment of rare diseases. A strong association between cancer and rare diseases was observed at the molecular level. Specifically, an overlap of approximately 60% was shown between 127 genes associated with cancer of many kinds and 2976 rare disease genes, and the same degree of overlap was also obtained when the analysis was conducted at the pathway level. By placing both gene lists mentioned above in a gene-gene network, over 95% gene pairs (one from each list) have two genes locating less than three genes apart in the network, indicating that cancer genes and rare disease genes likely involve similar biological processes. In addition, many drug targets for cancer were found to relate to rare diseases. The molecular level of association between cancer and rare diseases was further substantiated with existing clinical trial data and literature review. In summary, we ranked the rare disease classes by their potential to be treated with oncologic drugs. The study demonstrated that anticancer drugs are potential sources for the treatment of rare diseases, and the proposed framework offers an opportunity to identify potential therapeutics from cancer research for use in rare diseases.
AR-BIC-5

Landscape of circRNA Candidates Across 11 Organs and 4 Developmental Stages in Fischer 344 Rat
Binsheng Gong, Joshua Xu and Weida Tong

Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, AR

Circular RNA (circRNA) is a class of endogenous noncoding RNAs and has attracted great attention due to their potential biological function as regulators of microRNAs as recently reported in some studies. The next-generation sequencing technologies and novel bioinformatics approaches enable the detection of circRNAs in many species. Thousands of novel circRNA candidates have been revealed in mammalians as well as nematode. This study provides an overview of circRNA candidates detected through an RNA-seq dataset across 11 organs of Fischer 344 rats from 4 developmental stages. The induction of circRNA candidates displays clear organ-specific patterns and gender differences for some organs. Liver and muscle have the lowest numbers of circRNA candidates and brain has the most and the pattern was also observed for expressed genes and transcripts in our previous study. Among the 1,793 parental genes, only 58 were detected with backspliced junctions in all eleven organs and63 detected in ten organs but absent in one organ. 333 genes were detected with backspliced junctions only in one organ. The overlap of the induced circRNAs between male and female are less than 50% in each non-sexual organ, except for brain with an up to 67% concordance observed in aged rats. A trend of increase in circRNA candidates along the four developmental stages was observed in brain andliver for both sexes. In contrast, there is a drop in circRNA candidates in thymus for aged rats of both sexes. The number of circRNA candidates was stable in heart and lung. In the sex organs, the number of circRNA candidates remained stable across the aging points in Uterus, increased in the younger ages (Juvenile through Adult) in thymus and then significantly dropped for aged rats. Further knowledge of circRNA candidates in rat will undoubtedly advance the study of drug toxicity at the RNA regulation level.
AR-BIC-6

The Largest Reference Drug List Ranked by the Risk for Developing Drug-Induced Liver Injury in Humans
Minjun Chen (1), Ayako Suzuki (2), Shraddha Thakkar (1), Ke Yu (1), ChuChu Hu (1), and Weida Tong (1)

(1) Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, AR; (2) Department of Medicine, University of Arkansas for Medical Sciences, Little Rock, AR

Recently, personalized medicine has received a great attention to improve safety and effectiveness in drug development. Personalized medicine aims to provide medical treatment that is tailored to the patient’s characteristics such as genomic biomarkers, disease history, etc., so that the benefit of treatment can be optimized. Subpopulations identification is to divide patients into several different subgroups where each subgroup corresponds to an optimal treatment. For two subgroups, traditionally multivariate Cox proportional hazards model is fitted and used to calculate the risk score when outcome is survival time endpoint. Median is commonly chosen as the cutoff value to separate patients. Here we propose a novel tree-based method that adopts the algorithm of relative risk trees to identify subgroup patients. After growing a relative risk tree, we apply ??-means clustering to group the terminal nodes based on the averaged covariates. We adopt an ensemble Bagging method to improve the performance of a single tree since it is well known that the performance of a single tree is quite unstable. A simulation study is conducted to compare the performance between our proposed method and the multivariate Cox model. The applications of our proposed method to three public cancer data sets are also conducted for illustration.
AR-BIC-7

Association of Age at Menarche with Composition and Diversity of Gut Microbiota Among Women in the Twins UK Study
Yan Wang (1), Robert Delongchamp (2), Philip Williams (1), Mohammed El Faramawi (2), Mohammed Orloff (2), Galina Glazko (3), Yasir Rahmatallah (3), Jordana Bell (4), Tim Spector (4) and Barbara Fuhrman (2)

(1) Joint Bioinformatics Program UALR/UAMS, Little Rock, AR; (2) Department of Epidemiology, UAMS, Little Rock, AR; (3) Department of Biomedical Informatics, UAMS, Little Rock, AR; (4) Department of Twin Research and Genetic Epidemiology, King’s College London, UK

Early menarche is associated with increased risks of cardiovascular disease (CVD) incidence and mortality. We have hypothesized that this association results from a shared cause; alterations in the gut microbiome. We tested this hypothesis in a study of 908 female adult twins (542 dizygotic and 366 monozygotic) drawn from the TwinsUK registry. Age at menarche was self-reported. Microbial amplicons from the V3-V4 hypervariable regions of the 16S-rRNA gene were sequenced from fecal samples. Demultiplexed RNA sequences were downloaded from the European Bioinformatics Institute (EBI). We carried out quality filtering and open-reference Operational Taxonomic Unit (OTU) picking using QIIME. Sequences are aligned against the reference sequences of the Green Genes database. Sequences that fail the alignment are clustered de novo and cluster centroids are chosen as new reference sequences. Alpha diversity was measured in samples rarefied to a sequencing depth of 8,136. Mixed model regression analysis was used to model the association of menarche with microbial diversity while adjusting for covariance in twinned measures and for the effects of potential confounders, including birth cohort and BMI. Compared to the second and the third quantiles, women in the first and last quantiles for age at menarche had reduced alpha diversity. Age at menarche and alpha diversity of the fecal microbiome each accounted for some variations in CVD risk factors. The mechanisms underlying the observed associations remain to be elucidated.
AR-BIC-11

Alternative Splicing in Environmental Stress Regulated Genes
Thomas J, Hamilton M, Ramegowda V, Srivastava S, Basu S, Reddy AS and Pereira A

Crop, Soil and Environmental Sciences, University of Arkansas, Fayetteville, AR;
Colorado State University, Fort Collins, CO

Alternative splicing (AS) expands the transcriptome of humans and other metazoans, and is responsible for many human diseases due to mis-regulation. Plants cope with environmental stresses by expanding the transcriptome through AS, primarily through the mechanism of intron retention. However, most bioinformatics tools developed to analyze human AS are based on exon skipping. To analyze AS in plants new tools have been implemented, specifically for intron retention.  To study the response of rice to environmental stress, the rice cultivar Nipponbare (reference genome) was treated to drought and well-watered conditions at vegetative flag-leaf (V4) and early reproductive (R3) stages; as well as high-temperature at reproductive (R3) and grain filling (R6) stages. The response of rice in response to drought and temperature stress was studied using RNA-Seq, followed by bioinformatics analyses to identify and quantify differentially expressed genes and splice junctions by Tophat/Cufflinks and Splicegrapher. Predicted AS isoforms were validated by experimental approaches, and drought responsive AS transcripts were quantified using isoform specific primers for qPCR. Characterization of the isoform structure of AS genes by amplification, cloning and sequencing of individual isoforms of specific genes of interest, using primers designed across unique splice junctions that are differentially expressed under stress. The AS data of regulatory and downstream genes was used to develop models of regulation of drought tolerance by AS under drought.
AR-BIC-20

Best Practice in Mining Meaningful Topics from Regulatory Textual Documents
Weizhong Zhao, Yijun Ding, James J. Chen, Weida Tong, Roger Perkins and Wen Zou

Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, U.S. Food and Drug Administration, Jefferson, AR

Probabilistic topic modeling offers a viable approach to structure huge textual document collections into latent topic themes to aid text mining. FDA lore describes drug applications arriving in eighteen wheelers. Today the agency handles vast digital textual information from submissions representing some 25% of U.S. GDP, and untold terabytes of information from post market surveillance. Where experts are too few or slow, the means to extract information germane to regulatory questions is paramount. Here we describe extensive sensitivity studies to determine best practices for generating effective topic models. To test effectiveness and validity of topic models, we constructed a ground truth data set from PubMed that contained some 40 health related themes including negative controls, and mixed it with a data set of unstructured documents. The most useful models, tuned to desired sensitivity versus specificity, require an iterative process wherein preprocessing steps, the type of topic modeling algorithm, and the algorithm’s model parameters are systematically varied. Models need to be compared with both qualitative, subjective assessments and quantitative, objective assessments, and care is required that Gibbs sampling in model estimation is sufficient to assure stable solutions. With a high quality model, documents can be rank-ordered in accordance with probability of being associated with complex regulatory query string, greatly lessoning text mining work. Importantly, topic models are agnostic about how words and documents are defined, and thus our findings are extensible to topic models where samples are defined as documents, and genes, proteins or their sequences are words.
AR-BIC-21

Formula Milk Alters Microbial Diversity and Impacts Immune Response in Porcine Neonatal Model
Manish K. Saraf (1,2), Anne K. Bowlin (1,2), Sree V. Chintapalli (1,2), Kartik Shankar (1,2), Tanya LeRoith (3), Martin J. Ronis (4), Thomas M. Badger (1,2) and Laxmi Yeruva (1,2)

(1) Arkansas Children’s Nutrition Center, Arkansas Children Hospital Research Institute; ( 2) The University of Arkansas for Medical Sciences, Department of Pediatrics, Little Rock, Arkansas; (3) Department of Biomedical Sciences & Pathobiology, Virginia-Maryland College of Veterinary Medicine; (4) Department of Pharmacology & Experimental Therapeutics, Louisiana State University Health Sciences Center, New Orleans, LA

Formula feeding in infants is associated with increased risk of upper respiratory tract infections, allergies and gut dysfunction possibly by compromising gut immune system. We hypothesized that formula feeding alters the host microbiome, resulting in morphology changes and modulation of large intestine immune response. It was tested in piglets fed formula diet or sow from postnatal day (PND) 2 to 21 (n=12/group). Principal component analyses of microbial data indicated clear separation of the formula fed group from the control sow group and more microbial diversity (p=0.01) was observed. Formula feeding showed 2- to 8-fold abundance of Bacteroidaceae, Porphyromondadaceae, Rokenellaceae, Odoribacteraceae within Bacteroides phylum at family level in comparison to sow group. However, in the sow fed group 5-fold higher Paraprevotellaceae was observed (p<0.01). In Firmicutes only Streptococcus spp was 5-fold higher in formula fed piglets (p<0.05). Furthermore, 11-fold higher Gammaproteobacteria and 20 folder higher Verrucomicrabae was observed in formula fed pigelts (p<0.01) in comparison to sow-fed, indicating that formula-fed piglets show more microbial diversity than sow fed piglets in distal colon. Cytokine analyses showed 1.5- to 2.0-fold increases (p<0.05) in gene expression of BMP4, CCL21, CCL25, CSF3, VEGF-A and 1.5- to 4-fold (p<0.05) decreases in CXCL-11, IL-27 in PC and DC in formula group, suggesting the participation of specific cytokines in alteration of mucosal barrier and activation of gut-associated immune response. A significant increase in crypt density (p<0.05) was noticed in PC of formula-fed group, suggesting alterations in colon morphology. In summary, formula diet-driven microbiome changes accompanied alterations in colon crypt density and cytokine response. (USDA-ARS Project 6026-51000-010-05S).
AR-BIC-22