DTMiner: identification of potential disease targets through biomedical literature mining.

MOTIVATION: Biomedical researchers often search through massive catalogues of literature to look for potential relationships between genes and diseases. Given the rapid growth of biomedical literature, automatic relation extraction, a crucial technology in biomedical literature mining, has shown great potential to support research of gene-related diseases. Existing work in this field has produced datasets that are limited both in scale and accuracy. RESULTS: In this study, we propose a reliable and efficient framework that takes large biomedical literature repositories as inputs, identifies credible relationships between diseases and genes, and presents possible genes related to a given disease and possible diseases related to a given gene. The framework incorporates name entity recognition (NER), which identifies occurrences of genes and diseases in texts, association detection whereby we extract and evaluate features from gene-disease pairs, and ranking algorithms that estimate how closely the pairs are related. The F1-score of the NER phase is 0.87, which is higher than existing studies. The association detection phase takes drastically less time than previous work while maintaining a comparable F1-score of 0.86. The end-to-end result achieves a 0.259 F1-score for the top 50 genes associated with a disease, which performs better than previous work. In addition, we released a web service for public use of the dataset. AVAILABILITY AND IMPLEMENTATION: The implementation of the proposed algorithms is publicly available at http://gdr-web.rwebox.com/public_html/index.php?page=download.php The web service is available at http://gdr-web.rwebox.com/public_html/index.php CONTACT: jenny.wei@astrazeneca.com or kzhu@cs.sjtu.edu.cn Supplementary information: Supplementary data are available at Bioinformatics online.

An evaluation and recommendation of the optimal methodologies to detect RET gene rearrangements in papillary thyroid carcinoma.

To recommend a reliable and clinically realistic RET/PTC rearrangement detection assay for papillary thyroid carcinoma (PTC), we compared multiplex quantitative polymerase chain reaction (qPCR), fluorescence in situ hybridization (FISH), and immunohistochemistry (IHC). RET/PTC rearrangement was detected using either RET break-apart FISH followed by multicolor FISH to confirm CCDC6/RET or NCOA4/RET fusions, or by multiplex qPCR to detect 14 RET/PTC subtypes with simultaneous RET mRNA expression. RET protein expression was detected by IHC. The specificity and sensitivity of multiplex qPCR and IHC were calculated using break-apart FISH as a reference. Among 73 PTC patients with sufficient tissue available for FISH and multiplex qPCR, 10 cases were defined as RET/PTC positive by both assays, including eight CCDC6/RET and two NCOA4/RET fusions with relatively high RET mRNA. In addition, multiplex qPCR identified another two CCDC6/RET fusion positive cases, but with low RET mRNA expression. IHC staining identified 11 RET positive cases among 39 patients with available samples. In comparison to FISH, multiplex qPCR displayed 100% sensitivity and 97% specificity to detect RET/PTC fusions, while IHC was neither sensitive nor specific. Our data reveal that both multiplex qPCR and FISH assays are equally applicable for detection of RET/PTC rearrangements. Break-apart FISH methodology is highly recommended for the wider screening of RET rearrangements (regardless of partner genes), while multiplex qPCR is preferred to identify all known fusion types using one assay, provided mRNA expression is also measured. IHC analysis could potentially provide an additional method of fusion detection dependent on further optimization of assay conditions and scoring cutoffs.

A forest-based approach to identifying gene and gene gene interactions.

Multiple genes, gene-by-gene interactions, and gene-by-environment interactions are believed to underlie most complex diseases. However, such interactions are difficult to identify. Although there have been recent successes in identifying genetic variants for complex diseases, it still remains difficult to identify gene-gene and gene-environment interactions. To overcome this difficulty, we propose a forest-based approach and a concept of variable importance. The proposed approach is demonstrated by simulation study for its validity and illustrated by a real data analysis for its use. Analyses of both real data and simulated data based on published genetic models show the effectiveness of our approach. For example, our analysis of a published data set on age-related macular degeneration (AMD) not only confirmed a known genetic variant (P value = 2E-6) for AMD, but also revealed an unreported haplotype surrounding single-nucleotide polymorphism (SNP) rs10272438 on chromosome 7 that was significantly associated with AMD (P value = 0.0024). These significance levels are obtained after the consideration for a large number of SNPs. Thus, the importance of this work is twofold: it proposes a powerful and flexible method to identify high-risk haplotypes and their interactions and reveals a potentially protective variant for AMD.

Ultraspecific probes for high throughput HLA typing.

BACKGROUND: The variations within an individual's HLA (Human Leukocyte Antigen) genes have been linked to many immunological events, e.g. susceptibility to disease, response to vaccines, and the success of blood, tissue, and organ transplants. Although the microarray format has the potential to achieve high-resolution typing, this has yet to be attained due to inefficiencies of current probe design strategies. RESULTS: We present a novel three-step approach for the design of high-throughput microarray assays for HLA typing. This approach first selects sequences containing the SNPs present in all alleles of the locus of interest and next calculates the number of base changes necessary to convert a candidate probe sequences to the closest subsequence within the set of sequences that are likely to be present in the sample including the remainder of the human genome in order to identify those candidate probes which are "ultraspecific" for the allele of interest. Due to the high specificity of these sequences, it is possible that preliminary steps such as PCR amplification are no longer necessary. Lastly, the minimum number of these ultraspecific probes is selected such that the highest resolution typing can be achieved for the minimal cost of production. As an example, an array was designed and in silico results were obtained for typing of the HLA-B locus. CONCLUSION: The assay presented here provides a higher resolution than has previously been developed and includes more alleles than previously considered. Based upon the in silico and preliminary experimental results, we believe that the proposed approach can be readily applied to any highly polymorphic gene system.

LOT: a tool for linkage analysis of ordinal traits for pedigree data.

SUMMARY: Existing linkage-analysis methods address binary or quantitative traits. However, many complex diseases and human conditions, particularly behavioral disorders, are rated on ordinal scales. Herein, we introduce, LOT, a tool that performs linkage analysis of ordinal traits for pedigree data. It implements a latent-variable proportional-odds logistic model that relates inheritance patterns to the distribution of the ordinal trait. The likelihood-ratio test is used for testing evidence of linkage. AVAILABILITY: The LOT program is available for download at http://c2s2.yale.edu/software/LOT/

Analysis of twin data using SAS.

SUMMARY: Twin studies are essential for assessing disease inheritance. Data generated from twin studies are traditionally analyzed using specialized computational programs. For many researchers, especially those who are new to twin studies, understanding and using those specialized computational programs can be a daunting task. Given that SAS (Statistical Analysis Software) is the most popular software for statistical analysis, we suggest that the use of SAS procedures for twin data may be a helpful alternative and demonstrate that we can obtain similar results from SAS to those produced by specialized computational programs. This numerical validation is practically useful, because a natural concern with general statistical software is whether it can deal with data that are generated from special study designs such as twin studies and if it can test a particular hypothesis. We concluded through our extensive simulation that SAS procedures can be used easily as a very convenient alternative to specialized programs for twin data analysis.

Layout-aware information extraction from semi-structured medical images.

Textual information embedded in the medical image contains rich structured information about the medical condition of a patient. This paper aims at extracting structured textual information from semi-structured medical images. Given the recognized text spans of an image preprocessed by optical character recognition (OCR), due to the spatial discontinuity of texts spans as well as potential errors brought by OCR, the structured information extraction becomes more challenging. In this paper, we propose a domain-specific language, called ODL, which allows users to describe the value and layout of text data contained in the images. Based on the value and spatial constraints described in ODL, the ODL parser associates values found in the image with the data structure in the ODL description, while conforming to the aforementioned constraints. We conduct experiments on a dataset consisting of real medical images, our ODL parser consistently outperforms existing approaches in terms of extraction accuracy, which shows the better tolerance of incorrectly recognized texts, and positional variances between images. This accuracy can be further improved by learning from a few manual corrections.

Prospective study revealed prognostic significance of responses in leptomeningeal metastasis and clinical value of cerebrospinal fluid-based liquid biopsy.

OBJECTIVE: Leptomeningeal metastasis (LM) secondary to non-small cell lung cancer (NSCLC) is a devastating complication associated with poor prognosis. Diagnosis and assessment of responses in LM have been challenging due to limitation of traditional imaging tools and lack of standard evaluation criteria until very recently. To bridge this gap, we conducted the first prospective, observational study in cytologically diagnosed NSCLC-LM patients (NCT02803619). PATIENTS AND METHODS: A total of 49 NSCLC-LM patients were enrolled. LM responses were evaluated with a composite endpoint integrating neurological symptoms, cerebrospinal fluid (CSF) parameters and central nervous system (CNS) imaging. Primary outcome was overall survival (OS) after diagnosis of LM. Exploratory endpoint was the association between OS and prognostic factors. Primary tumor and CSF samples were collected for biomarker analysis. RESULTS: 93.9% of the cohort carried oncogenic drivers, and 85.7% harbored EGFR activating mutations. Median OS since LM diagnosis of the overall population was 9.7 months. EGFR mutant LM patients had a longer survival compared with wildtype ones. LM clinical responses assessed by the composite endpoint showed significant correlation with OS. Status of EGFR activating mutations was highly concordant between primary tumor and CSF. T790 M occurrence in CNS lesions was relatively rare and associated with intracranial exposure level of EGFR-TKIs. CONCLUSION: Our results supported the composite endpoint for objective response evaluation of LM was valid, suggested LM outweighed peripheral lesions on the impact to patient survival, and emphasized the urge and promise of development of CNS-penetrant targeted therapies to improve clinical outcome of NSCLC-LM patients.

Detection of EGFR mutations in plasma circulating tumour DNA as a selection criterion for first-line gefitinib treatment in patients with advanced lung adenocarcinoma (BENEFIT): a phase 2, single-arm, multicentre clinical trial.

BACKGROUND: Detection of EGFR mutations in tumour tissue is the gold-standard approach to ascertain if a patient will benefit from treatment with an EGFR tyrosine kinase inhibitor. However, if tissue is scant, another strategy is to use circulating tumour DNA (ctDNA), but this method needs validation in clinical trials. We did a prospective clinical trial to assess ctDNA-based EGFR mutation detection as a selection criterion for patients with lung adenocarcinoma receiving gefitinib as first-line treatment. METHODS: BENEFIT is a multicentre, single-arm, phase 2 clinical trial at 15 centres in China. Patients aged 18-75 years with stage IV metastatic lung adenocarcinoma and EGFR mutations detected in ctDNA were given oral gefitinib 250 mg once daily as first-line treatment. The primary endpoint was the proportion achieving an objective response. Secondary endpoints included median progression-free survival and safety. Next-generation sequencing (NGS) of a 168-gene panel was used for genetic analysis of baseline blood samples. The primary efficacy analysis was done by intention to treat in patients who had at least one post-baseline tumour assessment. The safety analysis was done in all patients who received at least one dose of study treatment. This trial is registered with ClinicalTrials.gov, number NCT02282267. FINDINGS: Between Dec 25, 2014, and Jan 16, 2016, 426 patients were screened for the trial, of whom 188 with EGFR mutations in ctDNA were enrolled and received gefitinib. 183 patients had one or more post-baseline tumour assessment and were included in the primary efficacy analysis. Median follow-up was 14·5 months (IQR 12·2-16·5). At the time of data cutoff (Jan 31, 2017), 152 patients had progressive disease or had died. The proportion achieving an objective response was 72·1% (95% CI 65·0-78·5). Median progression-free survival was 9·5 months (95% CI 9·07-11·04). Of 167 patients with available blood samples, 147 (88%) showed clearance of EGFR mutations in ctDNA at week 8, and median progression-free survival was longer for these patients than for the 20 patients whose EGFR mutations persisted at week 8 (11·0 months [95% CI 9·43-12·85] vs 2·1 months [1·81-3·65]; hazard ratio [HR] 0·14, 95% CI 0·08-0·23; p<0·0001). From baseline NGS data in 179 patients, we identified three subgroups of patients: those with EGFR mutations only (n=58), those with mutations in EGFR and tumour-suppressor genes (n=97), and those with mutations in EGFR and oncogenes (n=24). Corresponding median progression-free survival in these subgroups was 13·2 months (95% CI 11·5-15·0), 9·3 months (7·6-11·0), and 4·7 months (1·9-9·3), respectively (EGFR mutations only vs mutations in EGFR and tumour-suppressor genes, HR 1·78, 95% CI 1·23-2·58; p=0·002; EGFR mutations only vs mutations in EGFR and oncogenes, 2·66, 1·58-4·49; p=0·0003). The most common grade 3 or 4 adverse events were hepatic function abnormalities (n=24). Serious adverse events were reported in 17 (9%) patients. No unexpected safety events for gefitinib were recorded. INTERPRETATION: Detection of EGFR mutations in ctDNA is an effective method to identify patients who might benefit from first-line gefitinib treatment. Further analyses of dynamic alterations of EGFR mutations and accompanying gene aberrances could predict resistance to gefitinib. FUNDING: Guangdong Association of Clinical Trials, AstraZeneca, National Natural Sciences Foundation Key Programme, and National Key Research and Development Programme of China.

PD-L1 expression and its relationship with oncogenic drivers in non-small cell lung cancer (NSCLC).

In order to explore the potential patient population who could benefit from anti PD-1/PD-L1 mono or combination therapies, this study aimed to profile a panel of immunotherapy related biomarkers (PD-1, PD-L1, CTLA-4 and CD8) and targeted therapy biomarkers (EGFR, KRAS, ALK, ROS1 and MET) in NSCLC.Tumor samples from 297 NSCLC patients, including 156 adenocarcinomas (AD) and 129 squamous cell carcinomas (SCC), were analyzed using immunohistochemistry, immunofluorescence, sequencing and fluorescence in situ hybridization.43.1% of NSCLC patients had PD-L1 positive staining on ≥ 5% tumor cells (TC). Furthermore, dual color immunofluorescence revealed that the majority of PD-L1/CD8 dual positive tumor infiltrating lymphocytes (TIL) had infiltrated into the tumor core. Finally, combined analysis of all eight biomarkers showed that tumor PD-L1 positivity overlapped with known alterations in NSCLC oncogenic tumor drivers in 26% of SCC and 76% of AD samples.Our illustration of the eight biomarkers' overlap provides an intuitive overview of NSCLC for personalized therapeutic strategies using anti-PD-1/PD-L1 immune therapies, either as single agents, or in combination with targeted therapies. For the first time, we also report that PD-L1 and CD8 dual positive TILs are predominantly located within the tumor core.

Data-Driven Information Extraction from Chinese Electronic Medical Records.

OBJECTIVE: This study aims to propose a data-driven framework that takes unstructured free text narratives in Chinese Electronic Medical Records (EMRs) as input and converts them into structured time-event-description triples, where the description is either an elaboration or an outcome of the medical event. MATERIALS AND METHODS: Our framework uses a hybrid approach. It consists of constructing cross-domain core medical lexica, an unsupervised, iterative algorithm to accrue more accurate terms into the lexica, rules to address Chinese writing conventions and temporal descriptors, and a Support Vector Machine (SVM) algorithm that innovatively utilizes Normalized Google Distance (NGD) to estimate the correlation between medical events and their descriptions. RESULTS: The effectiveness of the framework was demonstrated with a dataset of 24,817 de-identified Chinese EMRs. The cross-domain medical lexica were capable of recognizing terms with an F1-score of 0.896. 98.5% of recorded medical events were linked to temporal descriptors. The NGD SVM description-event matching achieved an F1-score of 0.874. The end-to-end time-event-description extraction of our framework achieved an F1-score of 0.846. DISCUSSION: In terms of named entity recognition, the proposed framework outperforms state-of-the-art supervised learning algorithms (F1-score: 0.896 vs. 0.886). In event-description association, the NGD SVM is superior to SVM using only local context and semantic features (F1-score: 0.874 vs. 0.838). CONCLUSIONS: The framework is data-driven, weakly supervised, and robust against the variations and noises that tend to occur in a large corpus. It addresses Chinese medical writing conventions and variations in writing styles through patterns used for discovering new terms and rules for updating the lexica.

Intra-Tumoral Heterogeneity of HER2, FGFR2, cMET and ATM in Gastric Cancer: Optimizing Personalized Healthcare through Innovative Pathological and Statistical Analysis.

Current drug development efforts on gastric cancer are directed against several molecular targets driving the growth of this neoplasm. Intra-tumoral biomarker heterogeneity however, commonly observed in gastric cancer, could lead to biased selection of patients. MET, ATM, FGFR2, and HER2 were profiled on gastric cancer biopsy samples. An innovative pathological assessment was performed through scoring of individual biopsies against whole biopsies from a single patient to enable heterogeneity evaluation. Following this, false negative risks for each biomarker were estimated in silico. 166 gastric cancer cases with multiple biopsies from single patients were collected from Shanghai Renji Hospital. Following pre-set criteria, 56 ~ 78% cases showed low, 15 ~ 35% showed medium and 0 ~ 11% showed high heterogeneity within the biomarkers profiled. If 3 biopsies were collected from a single patient, the false negative risk for detection of the biomarkers was close to 5% (exception for FGFR2: 12.2%). When 6 biopsies were collected, the false negative risk approached 0%. Our study demonstrates the benefit of multiple biopsy sampling when considering personalized healthcare biomarker strategy, and provides an example to address the challenge of intra-tumoral biomarker heterogeneity using alternative pathological assessment and statistical methods.

Patient-Derived Gastric Carcinoma Xenograft Mouse Models Faithfully Represent Human Tumor Molecular Diversity.

Patient-derived cancer xenografts (PDCX) generally represent more reliable models of human disease in which to evaluate a potential drugs preclinical efficacy. However to date, only a few patient-derived gastric cancer xenograft (PDGCX) models have been reported. In this study, we aimed to establish additional PDGCX models and to evaluate whether these models accurately reflected the histological and genetic diversities of the corresponding patient tumors. By engrafting fresh patient gastric cancer (GC) tissues into immune-compromised mice (SCID and/or nude mice), thirty two PDGCX models were established. Histological features were assessed by a qualified pathologist based on H&E staining. Genomic comparison was performed for several biomarkers including ERBB1, ERBB2, ERBB3, FGFR2, MET and PTEN. These biomarkers were profiled to assess gene copy number by fluorescent in situ hybridization (FISH) and/or protein expression by immunohistochemistry (IHC). All 32 PDGCX models retained the histological features of the corresponding human tumors. Furthermore, among the 32 models, 78% (25/32) highly expressed ERBB1 (EGFR), 22% (7/32) were ERBB2 (HER2) positive, 78% (25/32) showed ERBB3 (HER3) high expression, 66% (21/32) lost PTEN expression, 3% (1/32) harbored FGFR2 amplification, 41% (13/32) were positive for MET expression and 16% (5/32) were MET gene amplified. Between the PDGCX models and their parental tumors, a high degree of similarity was observed for FGFR2 and MET gene amplification, and also for ERBB2 status (agreement rate = 94~100%; kappa value = 0.81~1). Protein expression of PTEN and MET also showed moderate agreement (agreement rate = 78%; kappa value = 0.46~0.56), while ERBB1 and ERBB3 expression showed slight agreement (agreement rate = 59~75%; kappa value = 0.18~0.19). ERBB2 positivity, FGFR2 or MET gene amplification was all maintained until passage 12 in mice. The stability of the molecular profiles observed across subsequent passages within the individual models provides confidence in the utility and translational significance of these models for in vivo testing of personalized therapies.

Recruitment challenges of a multicenter randomized controlled varicocelectomy trial.

OBJECTIVE: To review reasons for suboptimal recruitment for a randomized controlled trial (RCT) of varicocelectomy versus intrauterine insemination (IUI) for treatment of male infertility and to suggest means for improving future study recruitment. DESIGN: Survey of Reproductive Medicine Network (RMN) participating sites. SETTING: Reproductive Medicine Network. PATIENT(S): None. INTERVENTION(S): None. MAIN OUTCOME MEASURE(S): Ascertain reasons for inadequate recruitment and suggest improvements for future varicocelectomy trails. RESULT(S): This study screened seven and enrolled three couples, with the first couple randomized on June 30, 2010. The study was subsequently stopped on March 30, 2011. The following themes were cited most frequently by sites and therefore determined to be most likely to have played a role in suboptimal recruitment: [1] men must be screened at the beginning of a couple's infertility evaluation, [2] inclusion of infertile women who had failed previous fertility interventions appeared to be associated with the couple's intolerance of a placebo arm, and [3] an apparent bias against the use of unstimulated IUI cycles indicated a prejudicial preference for surgical intervention in the male partner. CONCLUSION(S): Improved recruitment may be realized through screening infertile men as early as possible while minimizing study-related time commitments. Focused patient education may promote improved equipoise and acceptance of a placebo arm in male infertility studies. Creative approaches to implementing varicocelectomy trials must be considered in addition to having a network of motivated researchers who carry a high volume of possible study participants because very large numbers may need to be screened to complete the clinical trial enrollment.

Total testosterone assays in women with polycystic ovary syndrome: precision and correlation with hirsutism.

CONTEXT: There is no standardized assay of testosterone in women. Liquid chromatography mass spectrometry (LC/MS) has been proposed as the preferable assay by an Endocrine Society Position Statement. OBJECTIVE: The aim was to compare assay results from a direct RIA with two LC/MS. DESIGN AND SETTING: We conducted a blinded laboratory study including masked duplicate samples at three laboratories--two academic (University of Virginia, RIA; and Mayo Clinic, LC/MS) and one commercial (Quest, LC/MS). PARTICIPANTS AND INTERVENTIONS: Baseline testosterone levels from 596 women with PCOS who participated in a large, multicenter, randomized controlled infertility trial performed at academic health centers in the United States were run by varying assays, and results were compared. MAIN OUTCOME MEASURE: We measured assay precision and correlation and baseline Ferriman-Gallwey hirsutism scores. RESULTS: Median testosterone levels were highest with RIA. The correlations between the blinded samples that were run in duplicate were comparable. The correlation coefficient (CC) between LC/MS at Quest and Mayo was 0.83 [95% confidence interval (CI), 0.80-0.85], between RIA and LC/MS at Mayo was 0.79 (95% CI, 0.76-0.82), and between RIA and LC/MS at Quest was 0.67 (95% CI, 0.63-0.72). Interassay variation was highest at the lower levels of total testosterone (≤50 ng/dl). The CC for Quest LC/MS was significantly different from those derived from the other assays. We found similar correlations between total testosterone levels and hirsutism score with the RIA (CC=0.24), LC/MS at Mayo (CC=0.15), or Quest (CC=0.17). CONCLUSIONS: A testosterone RIA is comparable to LC/MS assays. There is significant variability between LC/MS assays and poor precision with all assays at low testosterone levels.

Detecting Genes and Gene-gene Interactions for Age-related Macular Degeneration with a Forest-based Approach.

Age-related macular degeneration (AMD) is a leading cause of vision loss in the elderly. Genetic mechanisms underlying AMD are complex. Understanding the etiology of AMD is important because of the significant health and social concerns. In this paper, we describe a forest-based approach to systematically identifying multiple genes, gene-gene interactions and gene-environment interactions underlying complex diseases in genomewide case-control studies and the application of this approach to a published data set on AMD. Our analysis not only confirmed two known haplotypes, ACTCCG (on chromosome 1 with a p-value of 1.98e-6) and TCTGGACGACA (on chromosome 7 with a p-value of 9.81e-3), but also revealed two novel haplotypes, GATAGT (on chromosome 5 with a p-value of 3.46e-3) and TCTTACGTAGA (on chromosome 12 with a p-value of 3.16e-2). Thus, the significance of this work is twofold. First, we propose a powerful and robust method to identify high-risk haplotypes and their interactions; second, we reveal potential genetic variants associated with AMD.

A genome-wide association analysis of Framingham Heart Study longitudinal data using multivariate adaptive splines.

The Framingham Heart Study is a well known longitudinal cohort study. In recent years, the community-based Framingham Heart Study has embarked on genome-wide association studies. In this paper, we present a Framingham Heart Study genome-wide analysis for fasting triglycerides trait in the Genetic Analysis Workshop16 Problem 2 using multivariate adaptive splines for the analysis of longitudinal data (MASAL). With MASAL, we are able to perform analysis of genome-wide data with longitudinal phenotypes and covariates, making it possible to identify genes, gene-gene, and gene-environment (including time) interactions associated with the trait of interest. We conducted a permutation test to assess the associations between MASAL selected markers and triglycerides trait and report significant gene-gene and gene-environment interaction effects on the trait of interest.

Memory management in genome-wide association studies.

Genome-wide association is a powerful tool for the identification of genes that underlie common diseases. Genome-wide association studies generate billions of genotypes and pose significant computational challenges for most users including limited computer memory. We applied a recently developed memory management tool to two analyses of North American Rheumatoid Arthritis Consortium studies and measured the performance in terms of central processing unit and memory usage. We conclude that our memory management approach is simple, efficient, and effective for genome-wide association studies.

Detecting significant single-nucleotide polymorphisms in a rheumatoid arthritis study using random forests.

Random forest is an efficient approach for investigating not only the effects of individual markers on a trait but also the effect of the interactions among the markers in genetic association studies. This approach is especially appealing for the analysis of genome-wide data, such as those obtained from gene expression/single-nucleotide polymorphism (SNP) array experiments in which the number of candidate genes/SNPs is vast. We applied this approach to the Genetic Analysis Workshop 16 Problem 1 data to identify SNPs that contribute to rheumatoid arthritis. The random forest computed a raw importance score for each SNP marker, where higher importance score suggests higher level of association between the marker and the trait. The significance level of the association was determined empirically by repeatedly reapplying the random forest on randomly generated data under the null hypothesis that no association exists between the markers and the trait. Using random forest, we were able to identify 228 significant SNPs (at the genome-wide significant level of 0.05) across the whole genome, over two-thirds of which are located on chromosome 6, especially clustered in the region of 6p21 containing the human leukocyte antigen (HLA) genes, such as gene HLA-DRB1 and HLA-DRA. Further analysis of this region indicates a strong association to the rheumatoid arthritis status.

Using Mutual Information to Discover Temporal Patterns in Gene Expression Data.

Finding relations among gene expressions involves the definition of the similarity between experimental data. A simplest similarity measure is the Correlation Coefficient. It is able to identify linear dependences only; moreover, is sensitive to experimental errors. An alternative measure, the Shannon Mutual Information (MI), is free from the above mentioned weaknesses. However, the calculation of MI for continuous variables from the finite number of experimental points, N, involves an ambiguity arising when one divides the range of values of the continuous variable into boxes. Then the distribution of experimental points among the boxes (and, therefore, MI) depends on the box size. An algorithm for the calculation of MI for continuous variables is proposed. We find the optimum box sizes for a given N from the condition of minimum entropy variation with respect to the change of the box sizes. We have applied this technique to the gene expression dataset from Stanford, containing microarray data at 18 time points from yeast Saccharomyces cerevisiae cultures (Spellman et al.,[3]). We calculated MI for all of the pairs of time points. The MI analysis allowed us to identify time patterns related to different biological processes in the cell.