Volume 3 Supplement 7
Detecting significant single-nucleotide polymorphisms in a rheumatoid arthritis study using random forests
© Wang et al; licensee BioMed Central Ltd. 2009
Published: 15 December 2009
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.
Rheumatoid arthritis (RA), a common autoimmune disease with pathological symptom of swelling and inflammation, may cause severe damage to diarthrodial joints and even movement disability . Approximately 1% of the adult populations in the world are affected by RA, and women are more vulnerable to this disease than men . Research has led to improvements in RA therapy, but the etiology of the pathophysiological and molecular mechanisms underlying RA is not well understood.
Random forest has the ability to investigate the effect of the interactions among the markers in genetic association studies. A random forest consists of many classification trees and is known to be more reliability than a single tree in classification. Thus, it has been widely used for gene expression and single-nucleotide polymorphism (SNP) data analyses [3–5]. Using a random forest, one can estimate the impact of each variable on prediction results by calculating a measurement called importance . For example, Sun et al. used random forest for the RA status prediction on the Genetic Analysis Workshop (GAW) 15 RA data including 5742 SNPs . Although no single SNP was significant, they achieved an appealing prediction performance with their top 500 SNPs ranked by importance index. Based on these results, Sun et al. were able to accurately predict RA patients with high sensitivity and specificity. This conclusion can be explained by the fact an importance score is based not only on the contribution of a single variable, but also on its interactions with other variables. The importance index is a better measurement of an "overall" association of the variable with the disease compared to other statistics, e.g., the allelic chi-square statistic, designed for single marker association tests. However, one shortcoming of the importance index is that it does not seem to have an apparent, simple asymptotic distribution. Thus, it is hard to find significant SNPs from the results of importance index, even when we know high-ranking SNPs should have a better chance of being significant than low-ranking SNPs. Similar approaches to identifying important genes have been proposed by Chen et al.  for genome-wide studies and by Rodenburg et al.  for microarray data. In this study, we address this issue by estimating empirical p-values for importance index using the North American Rheumatoid Arthritis Consortium (NARAC) SNP data. We also apply a permutation model with the null hypothesis of no association between SNPs and RA status. Using a random forest approach, we were able to detect SNPs with significant importance.
The GAW16 Problem 1 RA dataset comprises whole-genome data from NARAC. It contains 868 cases and 1,194 controls, and 545,080 SNP-genotype fields from the Illumina 550 k chip. In this work, we first consider all 531,689 autosomal SNPs, discarding the mitochondrial, X, Y, and pseudo-autosomal chromosomes. By removing those SNPs with excessive missing data (missing >20%), we analyzed a total of 530,959 autosomal SNPs.
Random forest is a combinational classifier proposed by Breiman  that includes multiple individual classification trees. A classification tree is a classifier that uses a graphic tree model. The prediction procedure is based on a recursive partitioning method . Compared to a single classification tree, a random forest is regarded as a more reliable and robust alternative. To build a random tree in the random forest, first, a bootstrap dataset is formed by sampling with replacement from a training set. The size of the bootstrap dataset is equal to the training dataset. The unsampled data is known as the out-of-bag data. Next, a random set of the variables at each node is selected, and then the tree is fully grown without pruning. Finally, the previous process is repeated to develop multiple individual random trees. The strength of random forest lies in the fact that the ensemble can achieve both low-bias and more accurate results than a single classification tree.
where N represents the number of trees in random forest.
The test procedure proceeds in three steps. First, we apply the random forest by using individual SNPs as features and the case/control disease status as outcome and calculate the raw importance score of each SNP. Here the number of trees used in random forests is set to 20,000. Because of the massive number of SNPs (530,959), the number of SNPs used to split on at each node is set to be = 728. So statistically every SNP can be sampled at each node, especially at the root node. Second, to evaluate the significance of raw importance scores for a single SNP in its association with the disease status, the status of case/control is permuted randomly in the whole dataset. Then we reapply the random forest method on the permuted dataset and recalculate the importance scores. The maximum importance score over all the SNPs is recorded. Finally, the second step is repeated 5,000 times. Under the null hypothesis of no association between SNPs and the disease, a distribution of the maximum importance of single SNP is obtained based on the 5,000 runs, which can be used to assess the significance level of raw importance score in the original dataset.
Numbers of significant SNPs in 22 autosomal chromosomes
Significant SNPs in chromosomes 9
Significant SNPs in chromosomes 19
In this paper, we adopted the random forest approach for RA study on GAW16 Problem 1 and used a permutation procedure to estimate the significant level. As previously reported, our analysis confirmed that the region in 6p21 of chromosome 6 contains an abundance of significant SNPs. Also we found that some of the significant SNPs are related to known genes such as HLA-DRA1 and BTNL2. Moreover, we also detected two regions on chromosomes 9 and 19 for which significant linkage signals were previously reported. These results demonstrate that random forest can be a useful tool in detecting markers and chromosome regions that are in linkage disequilibrium with the disease alleles.
List of abbreviations used
Genetic Analysis Workshop
North American Rheumatoid Arthritis Consortium
This research is supported in part by grants K02 DA017713, R01 DA016750, and T32 MH014235 from the National Institutes of Health. The Genetic Analysis Workshop is supported by NIH grant R01 GM31575.
This article has been published as part of BMC Proceedings Volume 3 Supplement 7, 2009: Genetic Analysis Workshop 16. The full contents of the supplement are available online at http://www.biomedcentral.com/1753-6561/3?issue=S7.
- Firestein GS: Evolving concepts of rheumatoid arthritis. Nature. 2003, 423: 356-361. 10.1038/nature01661.View ArticlePubMedGoogle Scholar
- Callahan LF, Pincus T: Mortality in the rheumatic diseases. Arthritis Care Res. 1995, 8: 229-241. 10.1002/art.1790080406.View ArticlePubMedGoogle Scholar
- Sun YV, Cai Z, Desai K, Lawrance R, Leff R, Jawaid A, Kardia SL, Yang H: Classification of rheumatoid arthritis status with candidate gene and genome-wide single-nucleotide polymorphisms using random forests. BMC Proc. 2007, 1 (suppl 1): S62-10.1186/1753-6561-1-s1-s62.PubMed CentralView ArticlePubMedGoogle Scholar
- Chen X, Liu CT, Zhang M, Zhang H: A forest-based approach to identifying gene and gene gene interactions. Proc Natl Acad Sci USA. 2007, 104: 19199-19203. 10.1073/pnas.0709868104.PubMed CentralView ArticlePubMedGoogle Scholar
- Diaz-Uriarte R, Alvarez de Andres S: Gene selection and classification of microarray data using random forest. BMC Bioinformatics. 2006, 7: 3-10.1186/1471-2105-7-3.PubMed CentralView ArticlePubMedGoogle Scholar
- Breiman L: Random forests. Mach Learn. 2001, 45: 5-10.1023/A:1010933404324.View ArticleGoogle Scholar
- Rodenburg W, Heidema AG, Boer JM, Bovee-Oudenhoven IM, Feskens EJ, Mariman EC, Keijer J: A framework to identify physiological responses in microarray-based gene expression studies: selection and interpretation of biologically relevant genes. Physiol Genomics. 2008, 33: 78-90. 10.1152/physiolgenomics.00167.2007.View ArticlePubMedGoogle Scholar
- Zhang H, Singer B: Recursive Partitioning in the Health Sciences. 1999, New York, SpringerView ArticleGoogle Scholar
- Gregersen PK, Silver J, Winchester RJ: The shared epitope hypothesis. An approach to understanding the molecular genetics of susceptibility to rheumatoid arthritis. Arthritis Rheum. 1987, 30: 1205-1213. 10.1002/art.1780301102.View ArticlePubMedGoogle Scholar
- Newton JL, Harney SM, Wordsworth BP, Brown MA: A review of the MHC genetics of rheumatoid arthritis. Genes Immun. 2004, 5: 151-157. 10.1038/sj.gene.6364045.View ArticlePubMedGoogle Scholar
- Stammers M, Rowen L, Rhodes D, Trowsdale J, Beck S: BTL-II: a polymorphic locus with homology to the butyrophilin gene family, located at the border of the major histocompatibility complex class II and class III regions in human and mouse. Immunogenetics. 2000, 51: 373-382. 10.1007/s002510050633.View ArticlePubMedGoogle Scholar
- Orozco G, Eerligh P, Sánchez E, Zhernakova S, Roep BO, González-Gay MA, López-Nevot MA, Callejas JL, Hidalgo C, Pascual-Salcedo D, Balsa A, González-Escribano MF, Koeleman BP, Martín J: Analysis of a functional BTNL2 polymorphism in type 1 diabetes, rheumatoid arthritis, and systemic lupus erythematosus. Hum Immunol. 2005, 66: 1235-1241. 10.1016/j.humimm.2006.02.003.View ArticlePubMedGoogle Scholar
- Mandhyan DD, Kim-Howard X, Gaines M, Nath SK: A genome-wide ordered-subset linkage analysis for rheumatoid arthritis. BMC Proc. 2007, 1 (suppl 1): S101-10.1186/1753-6561-1-s1-s101.PubMed CentralView ArticlePubMedGoogle Scholar
- Chang M, Rowland CM, Garcia VE, Schrodi SJ, Catanese JJ, Helm-van Mil van der AH, Ardlie KG, Amos CI, Criswell LA, Kastner DL, Gregersen PK, Kurreeman FAS, Toes REM, Huizinga TWJ, Seldin MF, Begovich AB: A large-scale rheumatoid arthritis genetic study identifies association at chromosome 9q33.2. PLoS Genet. 2008, 4: e1000107-10.1371/journal.pgen.1000107.PubMed CentralView ArticlePubMedGoogle Scholar
- Thompson SD, Moroldo MB, Guyer L, Ryan M, Tombragel EM, Shear ES, Prahalad S, Sudman M, Keddache MA, Brown WM, Giannini EH, Langefeld CD, Rich SS, Nichols WC, Glass DN: A genome-wide scan for juvenile rheumatoid arthritis in affected sibpair families provides evidence of linkage. Arthritis Rheum. 2004, 50: 2920-2930. 10.1002/art.20425.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.