A statistical method for single sample analysis of HumanMethylation450 array data: genome-wide methylation analysis of patients with imprinting disorders
© Rezwan et al.; licensee BioMed Central. 2015
Received: 8 January 2015
Accepted: 6 April 2015
Published: 21 April 2015
The Illumina Infinium HumanMethylation450 BeadChip is an array-based technology for analysing DNA methylation at approximately 475,000 differentially methylated cytosines across the human genome. Hitherto, the array has been used for case-control studies, where sample numbers can be sufficient to yield statistically robust data on a genome-wide basis. We recently reported an informatic pipeline capable of yielding statistically and biologically significant results using only five cases, which expanded the use of this technology to rare disease studies. However, the clinical application of these technologies requires the ability to perform robust analysis of individual patients.
Here we report a novel informatic approach for methylation array analysis of single samples, using the Crawford-Howell t-test. We tested our approach on patients with ultra-rare imprinting disorders with aberrant DNA methylation at multiple locations across the genome, which was previously detected by targeted testing. However, array analysis outperformed targeted assays in three ways: it detected loci not normally analysed by targeted testing, detected methylation changes too subtle to detect by the targeted testing and reported broad and consistent methylation changes across genetic loci not captured by point testing.
This method has potential clinical utility for human disorders where DNA methylation change may be a biomarker of disease.
KeywordsMethylation Illumina HumanMethylation450 array Single case-control analysis Crawford-Howell t-test
Epigenetic modulation of gene expression is responsible for tissue specific and temporal changes across growth and development. The most widely studied of these epigenetic modifications is DNA methylation of 5-methylcytosine at CpG dinucleotides. Aberrations of DNA methylation are associated with a range of diseases, including imprinting disorders and cancer . Recent advances in technologies have made it possible to study the epigenetic changes associated with these diseases using robust genome-wide technologies including the Infinium HumanMethylation450 BeadChip (henceforward denoted the 450 k array; www.Illumina.com). The 450 k array measures the intensity of fluorescent signal from methylated and unmethylated probes at approximately 475,000 CpG dinucleotides across the genome, including CpG islands, promoters, gene bodies, intergenic regions and the majority of imprinted loci. These intensities are then used to calculate DNA methylation levels, with advantageous throughput, cost, coverage and technical consistency.
To date, many studies, utilising the 450 k array, have used case-control designs [2-6]. The limitation to the majority of these studies is that the bioinformatic analysis used requires a large number of cases and controls to obtain statistically significant results. Recently, we developed a novel informatic pipeline yielding statistically and biologically significant results using small case number analysis (case = 5) , which expanded the use of this technology to rare disease studies. However, the clinical application of these technologies requires the ability to perform robust analysis of individual patients.
Humans harbour approximately 100 known imprinted genes, characterised by the epigenetic control of gene expression, often through parent-of-origin-specific methylation that is applied in the germ line and conserved through subsequent development in all tissues. As yet, disruption of the methylation state at eight imprinted loci has been associated with imprinting disorders (Beckwith-Wiedemann syndrome (BWS; MIM #130659), Silver-Russell syndrome (SRS; MIM #180860), transient neonatal diabetes mellitus (TNDM; MIM #601410), Prader-Willi syndrome (PWS; MIM #176270), Angelman syndrome (AS; MIM #105830), matUPD14-like (Temple syndrome) and patUPD14-like (Wang-Kagami) syndromes and pseudohypoparathyroidism 1B (PHP-1B; MIM #103580)). Rare patients with multi-locus methylation disorders (MLMD) [8-11] form a uniquely informative group of samples that can be used to develop a sensitive and specific single sample 450 k array bioinformatic pipeline. Informatically, there are a number of approaches to single case analysis. A single normalised sample can be compared against a large sample group standardised in the same way . However, collecting a large normative sample can be both time-consuming and challenging . Another approach is to compare one or more tests to the performance of the same individuals by chi-square tests. However, the significant raw difference between different performances (or scores) can be diminished by comparison against control performance (or score) and vice versa. Alternatively, a single sample’s performance (or score) can be compared to that of a matched control group. Whereas the standardised method requires a large number of samples and intra-individual comparisons require assessment of two or more independent variables, the single case-control method requires only a moderate number of controls .
In single case-control analysis, the most common means of detecting significant differences is to convert the case’s score to a z-score using the control sample mean and standard deviation and referring the score to a table of areas under the normal curve . However, this might not accurately estimate the parameters if the control sample is large enough to assume that the mean and the standard deviation are used as population parameter rather than sample statistics . In many cases, the number of controls can be quite small (even smaller than 10). Therefore, it is logical to use a t-test method using the t-distribution. A number of studies used one-sample t-tests in their single case-control studies, and to date, several studies used Crawford and Howell’s t-test methods as mentioned in . The Weisberg t-test, for identifying outliers, is also capable of single sample analysis. The different t-tests will be briefly described in the following section.
However, all of these studies involved neuropsychological rather than 450 k array data. Here we demonstrate the effectiveness of a single case-control method for analysing 450 k array data from patients with multi-locus and single-locus imprinting disorders. Using 450 k array data from patients with known regions and severity of DNA hypomethylation, we were able to optimise our informatic approach: firstly, by comparison of various t-test methods and secondly, by varying the control group size to identify the smallest control size required to detect biologically and statistically significant changes in methylation at known regions of hypomethylation specific to each patient.
As mentioned earlier, we developed a pipeline for small sample size (n cases = 5) against large control groups, using patients with TND-MLMD and BWS-MLMD and broadly similar patterns of methylation change as determined by targeted testing. The pipeline applied a linear model as the statistical method, and CpGs were selected where they were hypomethylated compared with controls, with an adjusted P value < 1.33 × 10− 7 and M values between −1 and +1 (equivalent to 0.26 ≤ β ≤ 0.7) in normal controls, to enrich for the intermediate methylation consistent with the hemimethylation of genomic imprinting. We focused our attention on genes or DNA regions containing at minimum two CpGs within 2000 nucleotides. Using this approach, we detected 21 hypomethylated regions in the TND-MLMD and 34 regions in BWS-MLMD pooled samples , including regions of hypomethylation that were previously unknown and consistent with genomic imprinting. Targeted testing showed that some of these regions were not hypomethylated in all the samples. Therefore, though analysis of small case numbers vs. large control numbers could identify differential methylation robustly, it failed to identify patient-specific regions without targeted follow-up testing.
We applied single sample t-tests (Crawford-Howell, Weisberg and one-sample t-tests) instead of linear regression as the statistical method and modified the filtration criteria: hypomethylated DNA sequences with characteristics consistent with imprinting were selected as those containing a minimum of three consecutive CpGs within 2000 nucleotides with M value between −1 and +1 in normal controls and P value <0.05.
Selection of the CH t-test after comparative evaluation of t-test performance
Here we compared three types of t-tests (Crawford-Howell, one-sample and Weisberg t-tests) for their ability to identify known regions of differential methylation by single sample analysis while predicting less variability using a randomly selected control group size of 50 which was batch-matched, that is derived from the same batch of 450 k analyses as the patient DNA.
Therefore, both CH and Weisberg t-tests are capable of identifying regions of differential methylation with low false positive rate in single sample case-controls analysis. However, the CH t-test has the advantage of calculating effect size for single sample case-control studies, which is absent from the Weisberg t-test; therefore, we selected CH t-test for further analysis due to this and the more significant P values generated, and all further tests described here were performed using this method.
Detection of cardinal locations in MLMD patients using the single sample analysis
Detection of methylation disturbance at multiple locations using single sample analysis
Apart from the cardinal loci, the pipeline also detected additional significantly hypomethylated loci, including but not limited to those identified by targeted testing. Hypomethylation was detected at well-established imprinted loci including SNRPN, GNAS, MEST and GRB10, more recently identified loci including ZNF331, FAM50B, HM13, ERLIN2, LOC100130522, WRB and NHP2L1, and previously uninvestigated regions (such as SVOPL and MAFG; Additional file 1: Table S1).
Use of the single sample approach on ‘simple’ patients
In addition to the samples with MLMD, we analysed samples from patients with ‘simple’ imprinting disorders where targeted testing indicated that hypomethylation was restricted to the cardinal disease locus. In all four TND samples, the cardinal region of differential methylation was PLAGL1 as expected. For two samples, TND-SIMPLE 2 and 3, hypomethylation of five and one additional imprinted regions was identified respectively, characteristic of MLMDs (Additional file 1: Table S2). Additionally, one of the samples (TND-SIMPLE 2) had many novel regions of hypomethylation not previously associated with imprinted loci. One of these, GLP2R was also observed in TND-SIMPLE 3 as the only hypomethylated locus not associated with a known imprinting region. Likewise, all the BWS samples were hypomethylated at the cardinal locus KCNQ1OT1, but one sample (BWS-SIMPLE3) showed hypomethylation at multiple imprinting loci, characteristic of MLMD. This shows that 450 k-based analyses can detect methylation changes that may go undetected by the point determinations of targeted testing. (Additional file 1: Table S3).
Applying the single sample analysis to biological replicates
To assess whether the CH t-test detected false positives from control group variations, we processed one sample with two completely different batch-matched groups of 50 controls. The two tests respectively selected 205 and 184 hypomethylated CpG sites, with 170 sites in 12 regions in common (see Additional file 1: Table S4).
Determining the minimum number of controls
To assess the effect of control group size on detection of known regions of differential methylation, BWS-MLMD samples were analysed with varying numbers of controls (5, 10, 20, 30, 40 or 50) using the CH t-test. With 5 controls, no cardinal sites for BWS-MLMDs were detected. When control group size = 10, numerous regions of hypomethylation were identified (KCNQ1, PLAGL1, DIRAS3, MEST, GNAS, PEG3, NHP2L1, and PPIEL) though not WRB. With 20 controls all known regions of differential methylation were detected, the use of 30, 40 or 50 controls added little sensitivity. Similar results were obtained for TND-MLMD cases. In summary, 10 controls can produce statistically and biologically significant results, though 20 controls are preferable to obtain higher sensitivity. Additional file 1: Table S5 presents the sites found in TND-MLMD and BWS-MLMD samples using the varying control sizes using the CH t-test. No further significant improvement was observed using a larger control group (>50 controls), therefore, we restricted our maximum number of controls to 50.
Effect size calculation
In order to determine the effect of the CH t-test on the magnitude of differential methylation, the effect size for each sample was calculated against variable numbers of controls (5, 10, 20, 30, 40 or 50). For both TND-MLMD and BWS-MLMD samples, at the majority of differentially methylated regions, the effect sizes were similar irrespective of the number of controls. However, at some regions of differential methylation, the effect size was greater when control group size = 5 rather than ≥10. To determine the reliability of the effect size, point estimates and 95% confidence intervals of those effect sizes were calculated.
In general for TND-MLMD samples, the confidence interval was strikingly wider with 5 than 10 controls. For example, at the PLAGL1 region in TND-MLMD3, with 5 controls the effect size was −24.650, but the confidence interval was wide (−41.162 to −8.549), indicating this effect size as unreliable. With 40 controls, the effect size was much smaller (−13.875), but its confidence interval (−16.953 to −10.789) was tighter (Additional file 1: Table S6). With 20 controls, the effect size was intermediate −21.309, with confidence interval −28.033 to −14.574. The width of the confidence interval is attributed to the extreme hypomethylation of the PLAGL1 locus, which is a typical biological finding in TND but detrimental to effect size. For subtle changes in methylation, the effect sizes were much smaller, and use of 10, 20 or 30 controls resulted in large effect sizes with tight confidence intervals in both TND-MLMD and BWS-MLMD. Therefore, 20 controls appeared optimal for single sample analyses.
There are a number of motivations for developing this single sample case-control method for analysing Illumina 450 k methylation data. Firstly, the study population - patients with imprinting disorders - is small, and classic case-control studies would not yield statistically significant results. Secondly, individual patients have unique clinical features and unique epimutations and therefore require individual analysis to yield relevant epigenetic data with clinical utility. Thirdly, our former small-sample analysis approach  requires large control numbers to attain high statistical robustness, which is not always feasible for analysing single patients. Fourthly, use of large control batches would be prohibitively expensive if epigenomic array analysis is to be adopted as a pragmatic tool for epigenetic diagnosis of patients. Technical replication of the same sample in different batches, with different controls, clearly confirmed that our approach robustly detected statistically significant sites. The method also identified outlier samples, which is not possible for grouped case-control studies. The pipeline clearly identified one BWS-HIL patient with large abnormal DNA methylation variations (may be due to technical variations) in multiple locations, though these may be due to technical variation (Additional file 1: Table S1). It should be noted that the threshold P value of 0.05 is not as stringent as that used for case-control analyses (< 1.33 × 10− 7) but nonetheless does robustly identify imprinted loci.
The CH t-test method has the advantage of reporting not only probability of significant methylation changes but also the magnitude of the change by its effect size point estimate and confidence intervals. The power calculation interval shows the uncertainty of the point estimate of the effect size and its variation with the number of controls . Using this metric gave a concrete indication of the number of controls required to yield significant results.
The optimal number of controls for this approach was determined empirically, as the number of controls for which known imprinted loci were robustly detected. In broad terms, fewer than ten controls gave unreliable effect sizes (large confidence intervals), whereas control sizes of 10 and 20 gave improvements in confidence, and a modest additional improvement was achieved for >20 controls. We, therefore, suggest that 20 controls in the same batch are optimal for this approach, and using 10 controls is feasible in statistical terms. However, a requirement for large numbers of controls is not ideal for use in a diagnostic setting where cost is a consideration. We are currently attempting to identify robust methods for identifying methylation changes without the need for batch-matched controls. Though it is true that use of smaller numbers of controls runs the risk of violating the normality assumption, the effect of departure from normality is modest in case of the CH-test, as it is capable of controlling the Type I error rate [21,22]. While using large numbers of controls assures the normality of the distribution from the controls, in our empirical tests, we observed only incremental increases in statistical power with increase in control number above 20 controls.
We found the 450 k array to have unexpected benefits compared with targeted testing. Firstly, 450 k analysis is by definition an epigenome-wide approach and therefore detected DNA methylation variation at other loci not normally assessed in targeted testing for imprinting disorders. This expands of the scope of differentially methylated regions for future analysis. Secondly, 450 k data analysis was sensitive to subtle methylation changes at differentially methylated regions to the point where it detected variations that were undetected in targeted testing. Two cases that appeared by targeted testing to show ‘simple’ methylation changes (one TND and one BWS) were shown by 450 k array to have MLMD with subtle variations at several imprinted loci, which may be relevant to the clinical presentation of these individuals. This sensitivity probably stems from the fact that differentially methylated regions of imprinted genes frequently span tens or hundreds of CpG dinucleotides. While targeted testing is a single-point analysis, so a subtle variation may not be distinguishable from the normal range, whereas using the 450 k array a subtle variation may be reiterated many times sequentially, increasing its statistical robustness. Thirdly, differentially methylated regions are typified by dense clustering of CpG dinucleotides, and 450 k analysis reports on multiple CpGs in any given locus, and therefore, it gives information about the extent of methylation anomalies across a locus. This may offer novel information about the extent and effects of methylation changes across gene clusters.
An obvious limitation of 450 k-based analysis is that the array targets only a small percentage of potentially methylated cytosines in the genome; therefore, additional loci affected in these patients may remain undetected by this method. However, the disadvantage of incomplete coverage is offset by the advantages of cost and technical consistency.
450 k array-based analysis has not previously been used on patients with imprinting disorders, because their rarity and heterogeneity precluded the use of established case-control cohort studies. This is potentially very important for imprinting disorders, where standard diagnostic testing is fragmented, time-consuming and variably sensitive, and where clinically heterogeneous and overlapping features (for example pre- and post-natal growth dysregulation) can be associated with multiple epigenetic mutations, many of which are not included in current testing regimes. 450 k analysis offers potential for diagnosis of known imprinting disorders and for detection of novel patterns of methylation anomalies. This may lead to substantial improvements in the diagnostic rate and translational research for imprinting disorders, in the same way that genome-wide array analysis has advanced the clinical genetics of common diseases over the last fifteen years [23-25]. Intriguingly, methylation variation may also act as a biomarker of underlying genetic anomalies. It is well known that some deleterious genetic/genomic variations can be detected by means of consequent methylation changes: for example, FRAX triplet-repeat expansions cause promoter methylation and inactivation of the FRAX gene and Fragile X mental retardation , deletions and rearrangements of the IGF2 enhancer attenuate IGF2 expression with co-ordinate hypomethylation of promoter sequences  and genetic rearrangements in Lynch syndrome are detectable as epigenetic inactivation of MSH2 . We suggest that epigenome-wide DNA methylation analysis may be a powerful adjunct to genomic analysis, since it may indirectly indicate genomic variations that do not alter coding sequence but do alter gene expression.
Using the Crawford-Howell t-test in single sample case-control studies is a novel approach for analysing Illumina 450 k array methylation data. By this method, we identified statistically and biologically significant hypomethylation in individuals at both known and novel sites. We suggest that single sample analysis makes possible the use of the 450 k array as a translational research or diagnostic tool for human disorders associated with disturbance of DNA methylation.
Study and control populations
For this study, we selected patients with two imprinting disorders, Transient Neonatal Diabetes (TND) and Beckwith-Wiedemann Syndrome (BWS). These patients have been described previously, and their methylation levels determined at several imprinted loci by targeted testing [7,9,29]. In our recent study , five multi-locus methylation disorder patient samples from each clinically classified group (TND or BWS) were processed in separate batches with 245 and 221 anonymous healthy controls, respectively, from an unrelated study. In this study, we additionally included four TND and three BWS patients where targeted testing detected DNA hypomethylation only at the cardinal disease loci, with no known involvement of any other imprinted locus (denoted ‘simple’ BWS and TND cases). These samples were processed in a third batch with 63 anonymous healthy controls from an unrelated study. Batch-matched controls were chosen as the control group and randomly selected for each sample analysed in the single sample analysis pipeline.
To assess the methylation level in each sample, a standard workflow was followed. The DNA in each sample was extracted from the whole blood by the standard procedure described in , and DNA concentration was determined using PicoGreen dsDNA Quantitation Kit (Molecular Probes, Inc., OR, USA). One microgram of DNA was bisulfite-treated for converting cytosine to thymine using the EZ 96-DNA Methylation Kit (Zymo Research, CA, USA). Illumina Infinium HumanMethylation450 BeadChip (Illumina, Inc., CA, USA), which was processed following standard protocol , was used to estimate genome-wide DNA methylation. Multiple identical control samples were assigned to each batch to assess assay variability and control batch effects. The BeadChips were scanned by the BeadStation and the methylation levels, as beta (β) values, were extracted using the Methylation Module of GenomeStudio (version 2011.1). The methylation data were then pre-processed further, as described in the following section.
Single sample analysis pipeline
Pre-processing of the 450 k data first removes any CpG sites with missing values, followed by removal of any sample where >90% CpG sites have detection P value >0.05, and any CpG sites where >75% samples have detection P value >10−5. Probes on the X and Y chromosomes were removed to discard any sex bias within the samples. The beta-values were converted to logit transformed M values, and quantile normalisation was used to normalise signal intensities to reduce inter-array variation . Peak correction  was applied to correct differences between Infinium I and Infinium II type assays. No batch correction was required as each case and its corresponding controls were drawn from the same batch. Statistically significant differences between the pre-processed M values of cases and controls were determined using single sample t-tests.
Statistical tests for identifying significant differential methylation
To establish the optimal test for or single sample analysis, we compared the CH t-test method to other two t-tests, namely - one-sample and Weisberg t-tests.
Power calculations for single sample analysis
The t-value from the CH t-test shows the statistical significance of the difference between case and controls, whereas the effect size index shows the level of difference between them. Along with the point estimate of the effect size, an estimate interval should also be presented in the single sample analysis. The procedure used in this paper to measure the confidence interval of the point estimate of the effect size has been described previously in , and the calculation is further explained in .
To reduce false positive calls, we further filtered the results from the significant difference between case and controls groups by CH t-test and power calculation. To define sites that were hypomethylated in cases, we initially set the same stringent criteria as in : one-tailed P value (adjusted using false discovery rate) < 10−7 and M value between −1 and +1 in normal controls with the beta-differences smaller than zero (to select only hypomethylated loci). Genes containing at least three CpGs meeting these criteria within <2000 bp (base pair) were selected as candidate DMRs consistent with imprinting. However, when applied to single sample analysis, these criteria were too stringent to detect known differentially methylated regions. Therefore, for single sample analyses, we used a less stringent P value, which was calculated as described in Statistical tests for identifying significant differential methylation: significant methylation changes were therefore selected as those containing a minimum of three consecutive CpGs within 2000 nucleotides with M values between −1 and +1 in normal controls and P value <0.05.
Minimum number of controls
Using varying numbers of controls (5, 10, 20, 30, 40 or 50) we assessed the impact of control group size in detecting known regions of hypomethylation and changes in effect size and confidence interval in our single sample analysis.
The authors would like to thank Peter Henneman for his valuable discussions that improved the manuscript. We thank the High-Throughput Genomics Group at the Wellcome Trust Centre for Human Genetics (funded by Wellcome Trust grant reference 090532/Z/09/Z and MRC Hub grant G0900747 91070) for the generation of the methylation data.
This work was supported by the Medical Research Council, UK (MR/J000329/1 to FIR and LED). Funding for open access charge was supported by the Research Councils UK. The generation of the normal control population methylation data was supported by the National Institute of Allergy and Infectious Diseases under Award Number R01 AI091905-01 (PI: Wilfried Karmaus).
- Feinberg AP. Phenotypic plasticity and the epigenetics of human disease. Nature. 2007;447:433–40.View ArticlePubMedGoogle Scholar
- Yang BZ, Zhang H, Ge W, Weder N, Douglas-Palumberi H, Perepletchikova F, et al. Child abuse and epigenetic mechanisms of disease risk. Am J Prev Med. 2013;44:101–7.View ArticlePubMed CentralPubMedGoogle Scholar
- Shenker NS, Polidoro S, van Veldhoven K, Sacerdote C, Ricceri F, Birrell MA, et al. Epigenome-wide association study in the European Prospective Investigation into Cancer and Nutrition (EPIC-Turin) identifies novel genetic loci associated with smoking. Hum Mol Genet. 2013;22:843–51.View ArticlePubMedGoogle Scholar
- Moore K, McKnight AJ, Craig D, O’Neill F. Epigenome-wide association study for Parkinson’s disease. Neuromolecular Med. 2014;16:845–55.View ArticlePubMedGoogle Scholar
- Seow WJ, Kile ML, Baccarelli AA, Pan WC, Byun HM, Mostofa G, et al. Epigenome-wide DNA methylation changes with development of arsenic-induced skin lesions in Bangladesh: a case-control follow-up study. Environ Mol Mutagen. 2014;55:449–56.View ArticlePubMed CentralPubMedGoogle Scholar
- Abdolmaleky HM, Nohesara S, Ghadirivasfi M, Lambert AW, Ahmadkhaniha H, Ozturk S, et al. DNA hypermethylation of serotonin transporter gene promoter in drug naive patients with schizophrenia. Schizophr Res. 2014;152:373–80.View ArticlePubMedGoogle Scholar
- Docherty LE, Rezwan FI, Poole RL, Jagoe H, Lake H, Lockett GA, et al. Genome-wide DNA methylation analysis of patients with imprinting disorders identifies differentially methylated regions associated with novel candidate imprinted genes. J Med Genet. 2014;51:229–38.View ArticlePubMed CentralPubMedGoogle Scholar
- Mackay DJ, Callaway JL, Marks SM, White HE, Acerini CL, Boonen SE, et al. Hypomethylation of multiple imprinted loci in individuals with transient neonatal diabetes is associated with mutations in ZFP57. Nat Genet. 2008;40:949–51.View ArticlePubMedGoogle Scholar
- Bliek J, Verde G, Callaway J, Maas SM, De Crescenzo A, Sparago A, et al. Hypomethylation at multiple maternally methylated imprinted regions including PLAGL1 and GNAS loci in Beckwith-Wiedemann syndrome. Eur J Hum Genet. 2009;17:611–9.View ArticlePubMed CentralPubMedGoogle Scholar
- Azzi S, Rossignol S, Steunou V, Sas T, Thibaud N, Danton F, et al. Multilocus methylation analysis in a large cohort of 11p15-related foetal growth disorders (Russell Silver and Beckwith Wiedemann syndromes) reveals simultaneous loss of methylation at paternal and maternal imprinted loci. Hum Mol Genet. 2009;18:4724–33.View ArticlePubMedGoogle Scholar
- Eggermann T. Russell-Silver syndrome Imprinted genes and human disease. Am J Med Genet Part C Semin Med Genet. 2010;154c(3):355–64. doi:10.1002/ajmg.c.30274.View ArticlePubMedGoogle Scholar
- Crawford JR, Howell DC. Comparing an individual’s test score against norms derived from small samples. Clin Neuropsychol. 1998;12:5.View ArticleGoogle Scholar
- Crawford JR. Psychometric foundations of neuropsychological assessment. In: Laura H, Goldstein JE, editors. Clinical Neuropsychology: a practical guide to assessment and management for clinicians. 2nd ed. Chichester: Wiley; 2004.Google Scholar
- Howell DC. Statistical methods for psychology. 5th ed. Belmont, CA: Duxbury Press; 2002.Google Scholar
- Crawford JR, Garthwaite PH. Comparison of a single case to a control or normative sample in neuropsychology: development of a Bayesian approach. Cogn Neuropsychol. 2007;24:343–72.View ArticlePubMedGoogle Scholar
- Crawford JR, Garthwaite PH. Single-case research in neuropsychology: a comparison of five forms of t-test for comparing a case to controls. Cortex. 2012;48:1009–16.View ArticlePubMedGoogle Scholar
- Reinhold N, Markowitsch HJ. Emotion and consciousness in adolescent psychogenic amnesia. J Neuropsychol. 2007;1:53–64.View ArticlePubMedGoogle Scholar
- Vecera SP, Rizzo M. What are you looking at? Impaired ‘social attention’ following frontal-lobe damage. Neuropsychologia. 2004;42:1657–65.View ArticlePubMedGoogle Scholar
- Brand M, Kalbe E, Kracht LW, Riebel U, Munch J, Kessler J, et al. Organic and psychogenic factors leading to executive dysfunctions in a patient suffering from surgery of a colloid cyst of the Foramen of Monro. Neurocase. 2004;10:420–5.View ArticlePubMedGoogle Scholar
- Crawford JR, Garthwaite PH, Porter S. Point and interval estimates of effect sizes for the case-controls design in neuropsychology: rationale, methods, implementations, and proposed reporting standards. Cogn Neuropsychol. 2010;27:245–60.View ArticlePubMedGoogle Scholar
- Crawford JR, Garthwaite PH. Testing for suspected impairments and dissociations in single-case studies in neuropsychology: evaluation of alternatives using monte carlo simulations and revised tests for dissociations. Neuropsychology. 2005;19:318–31.View ArticlePubMedGoogle Scholar
- Crawford JR, Garthwaite PH, Azzalini A, Howell DC, Laws KR. Testing for a deficit in single-case studies: effects of departures from normality. Neuropsychologia. 2006;44:666–77.View ArticlePubMedGoogle Scholar
- de Vries BB, Pfundt R, Leisink M, Koolen DA, Vissers LE, Janssen IM, et al. Diagnostic genome profiling in mental retardation. Am J Hum Genet. 2005;77:606–16.View ArticlePubMed CentralPubMedGoogle Scholar
- Menten B, Maas N, Thienpont B, Buysse K, Vandesompele J, Melotte C, et al. Emerging patterns of cryptic chromosomal imbalance in patients with idiopathic mental retardation and multiple congenital anomalies: a new series of 140 patients and review of published reports. J Med Genet. 2006;43:625–33.View ArticlePubMed CentralPubMedGoogle Scholar
- Stankiewicz P, Beaudet AL. Use of array CGH in the evaluation of dysmorphology, malformations, developmental delay, and idiopathic mental retardation. Curr Opin Genet Dev. 2007;17:182–92.View ArticlePubMedGoogle Scholar
- Gerhardt J, Zaninovic N, Zhan Q, Madireddy A, Nolin SL, Ersalesi N, et al. Cis-acting DNA sequence at a replication origin promotes repeat expansion to fragile X full mutation. J Cell Biol. 2014;206:599–607.View ArticlePubMed CentralPubMedGoogle Scholar
- Gronskov K, Poole RL, Hahnemann JM, Thomson J, Tumer Z, Brondum-Nielsen K, et al. Deletions and rearrangements of the H19/IGF2 enhancer region in patients with Silver-Russell syndrome and growth retardation. J Med Genet. 2011;48:308–11.View ArticlePubMedGoogle Scholar
- Ligtenberg MJ, Kuiper RP, Chan TL, Goossens M, Hebeda KM, Voorendt M, et al. Heritable somatic methylation and inactivation of MSH2 in families with Lynch syndrome due to deletion of the 3′ exons of TACSTD1. Nat Genet. 2009;41:112–7.View ArticlePubMedGoogle Scholar
- Mackay DJ, Boonen SE, Clayton-Smith J, Goodship J, Hahnemann JM, Kant SG, et al. A maternal hypomethylation syndrome presenting as transient neonatal diabetes mellitus. Hum Genet. 2006;120:262–9.View ArticlePubMedGoogle Scholar
- Miller SA, Dykes DD, Polesky HF. A simple salting out procedure for extracting DNA from human nucleated cells. Nucleic Acids Res. 1988;16:1215.View ArticlePubMed CentralPubMedGoogle Scholar
- Bibikova M, Fan JB. GoldenGate assay for DNA methylation profiling. Methods Mol Biol. 2009;507:149–63.View ArticlePubMedGoogle Scholar
- Wang D, Yan L, Hu Q, Sucheston LE, Higgins MJ, Ambrosone CB, et al. IMA: an R package for high-throughput analysis of Illumina’s 450 K Infinium methylation data. Bioinformatics. 2012;28:729–30.View ArticlePubMed CentralPubMedGoogle Scholar
- Dempster EL, Pidsley R, Schalkwyk LC, Owens S, Georgiades A, Kane F, et al. Disease-associated epigenetic changes in monozygotic twins discordant for schizophrenia and bipolar disorder. Hum Mol Genet. 2011;20:4786–96.View ArticlePubMed CentralPubMedGoogle Scholar
- Dedeurwaerder S, Defrance M, Calonne E, Denis H, Sotiriou C, Fuks F. Evaluation of the Infinium Methylation 450 K technology. Epigenomics. 2011;3:771–84.View ArticlePubMedGoogle Scholar
- Crawford JR, Garthwaite PH. Investigation of the single case in neuropsychology: confidence limits on the abnormality of test scores and test score differences. Neuropsychologia. 2002;40:1196–208.View ArticlePubMedGoogle Scholar
- Weisberg S. Probability and mathematical statistics: applied linear regression. 2nd ed. New York: John Wiley & Sons; 1985.Google Scholar
- Sullivan GM, Feinn R. Using effect size-or why the p value is not enough. J Grad Med Educ. 2012;4:279–82.View ArticlePubMed CentralPubMedGoogle Scholar
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.