(Universal declaration of the Human Genome and Human Rights)

2.1.1 Global visualization by genome browsers

A sea is so large that even an experienced sailor needs the navigating tools of maps and compass to direct his exploration. Similarly, the deluge and richness of genomic information prohibit investigators from comprehension and analysis without the aids of computational tools. Computers are not only used to store, organize, search, analyze and present the genomic information, but also to execute adaptive computational models to illustrate the characteristics of the data. Visual presentations are crucial for conveying the rich context of genomic information in either the printed forms or the interactive, searchable forms called genome browsers. Here we introduce three commonly used diagrams for genomics. First, a karyogram (also known as an idiogram) is a global, bird’s eye view of the presentation of the 23 pairs of chromosomes and their annotations. For example, it can be used to present the tumor suppressor genes by showing their locations on the chromosome. One of the popular tools for producing a virtual karyogram is provided by the University of California, Santa Cruz (UCSC) Genome Bioinformatics site (Box 2.1).

Second, a genome browser is an integrative, interactive visualization tool that can present genomic information, chromosome by chromosome, with multiple resolutions of annotations. Several genome browsers are part of important bioinformatic portals, for example, the Ensemble and HapMap websites (Box 2.1). A genome browser can also be standalone software, for example, the freeware “GenomeBrowse” offered by Golden Helix Inc. A genome browser usually has an interactive, zooming in and zooming out interface to navigate to regions of interest and adjust to the right scale/resolution. This enables the visualization of large-scale contexts and small-scale details. It also has multiple tracks, which can each be designated to a special type of annotation such as the exon and intron locations, personal variants (to be discussed later), splicing patterns and functional annotations. The presentation of multiple tracks simultaneously offers an integrative view of complex information.

A genome browser is traditionally designed as a linear, horizontal interactive graph with multiple layers of annotation showing simultaneously as tracks. With the current requirement of presenting complex gene interactions and structural variations across chromosomes, a new type of genome browser has been designed with a circular shape with the chromosomes forming the circle. Annotations are then represented in the circular tracks, and the interactions of loci are shown as arcs within the circle. Recently, such a type of genome diagram and browser was constructed by the CIRCOS software (Box 2.1).

2.2.1 Genomic variants

A substantial portion of the human genome is identical amongst people; the differences are called variants. Many such variants are scattered in the human genome. They can be classified, based on their lengths (scales), as:

The collection of all variations found in the human population is called the human variome.

The single-base SNPs and point mutations are important classes of genomic variants, partly because of their sheer numbers. Large-scale human genome sequencing projects, such as the 1000 Genome Project and the uk10k Project (a 6000 vs. 4000 case vs. control study aiming to conduct the sequencing of the protein-coding regions), have increased the SNP count to 20 million (Pennisi, 2010). Both SNPs and point mutations have two (occasionally three) alleles: the more common one is called the major allele, while the less common one is called the minor allele. A SNP is defined as having a minor allele frequency (MAF) higher than 1% in a population. The International HapMap Project has achieved a valuable database that can be used to calculate allele frequencies across populations (The International HapMap Consortium, 2007; Altshuler et al., 2010). The HapMap Project is prioritized to collect common variants with MAF higher than 1%. The core of the database is a large SNP-by- s ubject matrix of genotypes. The Phase 3 collection contains genotypes of 1,600,000 SNPs of 1184 subjects in 11 populations. The database has many uses, such as the comparison of allele frequencies in different ethnic groups and the demonstration of genomic structure based on correlations between SNPs.

On average, there is 1 SNP in every 150 nucleotides (20 million SNPs scattered in a 3 billion nucleotide genome, see Table 2.1). Hence, SNPs offer dense genomic markers. Each SNP has a unique position in the genome and is assigned a unique identification (ID) by dbSNP, one of the most important resources of SNPs hosted by the National Center for Biotechnology Information (NCBI) site (Box 2.1).

In addition to SNPs and point mutations, short tandem repeats and interspersed repeats are commonly found in the human genome. Alu is a class of primate specific, short interspersed repeats, which constitutes approximately 10% of the human genome (Lander et al., 2001; Decerbo and Carmichael, 2005). Cellular roles of Alu have been obscure and it was once thought of as junk DNA (Muotri et al., 2007; Lin et al., 2008). Alu can be found in intergenic regions as well as the coding and non- coding regions of many genes in either the sense or antisense strand.

Recent studies on human genome analysis have revealed a common class of germline variants, the CNVs that belong to a more general class called structural variants. The genome is thought to be diploid where one copy is from the father and one from the mother. Yet, on the sub- microscopic scale, multiple genomic regions exhibit multiple elevated (e.g. above 3 copies) or reduced (e.g. 1 or 0 copies) numbers of copies, and are called CNVs. These are common, mostly harmless variants, although some have been linked to disease. The shortest length of a CNV is 1 kb based on its definition, yet many of them are longer than 100 kb. It is estimated that an average Asian person has 20 to 30 CNV loci in their genome, with an average size of 14.1 kb (Wang et al., 2007).

Cohorts of subjects need to be recruited to reveal the basic characteristics of genomic variants, such as total number and population frequency. The HapMap Project is a good example. In addition, we also need to find the genomic variants responsible for individual variability in disease susceptibility or drug response. This can also be achieved by genomic studies with adequate sample size, rigorous statistical measures, and unambiguous definition and recruitment of subjects with distinct clinical phenotypes.

Somatic mutations of various scales play significant roles in the etiology of many illnesses, particularly cancer. Most cancers originated from somatic cells transformed via a series of genetic or epigenetic modifications during the aging process (Luebeck, 2010; Stephens et al., 2011). These cells gradually exhibit many abnormal capabilities such as anti-apoptosis, tissue invasion and metastasis (Gupta and Massague, 2006). Somatic mutations of oncogenes or tumor suppressor genes could serve as important biomarkers in various stages of cancer biology. In terms of treatment, traditional chemotherapy agents have been designed against fast dividing cells. The more recent target therapies employ antagonists to various oncogenes directly. Cancer patients show varied responses, chemotherapy or targeted therapy alike, because their cancer cells are at different stages and with various features and capabilities. This is why the optimum treatment strategies differ. Studies of somatic mutations on targeted genes, oncogenes and tumor suppressor genes may be able to identify the best treatment strategy for each individual.

2.2.2 A multiscale tagging of your genome

Linkage disequilibrium (LD) is a unique feature among genomic variants. It refers to the association of genotypes between adjacent variants. It is a direct consequence of chromosomal recombination (also known as crossover), which occurs during meiosis for the production of germ cells. Where there are no chromosomal recombinations, all variants in one chromosome should be tightly associated to each other, as they are passed down together to future generations. A chromosomal recombination disrupts the association (and introduces independence) between variants at the two sides of the recombination spot. After several generations, distant variants are no longer associated due to many recombination hotspots between them. The adjacent variants still retain a certain degree of LD, particularly those between two recombination hotspots. LD of a pair of variants can be quantified by r² or D’ values. r² is a value between 0 and 1, while 0 indicates no association and 1 indicates high association. LD can be shown as LD-heat maps by the Haploview software or the SeattleSNPs site (Box 2.1).

LD is a cornerstone for the design of genotype vs. phenotype association studies. Since many variants are adjacent to the causative variant, their genotype may also be in LD with the causative genotype. In the results of many high density association studies, we have seen multiple adjacent SNPs associated to the clinical trait simultaneously. The concept of using adjacent variants to serve as surrogates can alleviate the burden of finding the exact causative variants directly. We can first detect the surrogates, then scrutinize the proximity region of the surrogate to find the real causative variant. In practice, this is realized by the selection of representative SNPs (known as tag SNPs) in the study design phase, assuming the true causative variant is in LD with the tag SNPs.

A variety of tagging software is available based on r² LD statistics (Box 2.1). For example, a tag SNP selection software, called the LDSelect method, has been implemented for data organization and analysis on the SeattleSNPs website (Carlson et al., 2004). LDSelect can ensure the selected tag SNPs have relatively low LD between each other, thereby minimizing the number of SNPs while maximizing their representativity. The FastTagger is a GNU GPL open source software (Liu et al., 2010). It implements two tagging algorithm, MultiTag and MMTagger. FastTagger comprise two consecutive steps:

Tagging is conventionally applied to a chromosomal region, with a fixed window size, due to computational limitations. By extending the concept of tagging from one layer into multiscale tagging, we can achieve the tagging of genome-wide variants. This can be achieved by a bottom-up layer construction approach (Figure 2.1). The entire set of genomic variants is defined as layer 0. We then tagged on top of layer 0, using a defined r² threshold and a window size, to produce layer 1 containing only the tagged variants. The tagging reduced the number of variants in layer 1 compared with layer 0. A hierarchical structure can be further constructed, layer by layer iteratively, with an increasing window size. Every layer comprises the tagging SNPs of the immediately lower layer. After a few iterations, the top layer is constructed, serving as a reasonable representation of genome-wide variants.

2.2.3 Haplotypes

The complex disease common variant (CDCV) hypothesis has been the underlying hypothesis of mainstream genome studies in the first decade of the 21st century (Goldstein, 2009). It influences many genome-wide association studies (GWAS) and early phases of the HapMap Project. This hypothesis assumes that complex disease is more likely to be caused by SNPs with higher MAFs (i.e. common alleles) than rarer SNPs and mutations. Therefore, higher priority is given to the searching, genotyping and disease association of common SNPs. But how about the rarer SNPs, mutations and other types of rare variants, which are also likely to be responsible for many diseases? Apart from using next generation sequencing (NGS) to capture all of them, an alternative way to accommodate them is by employing the concept of haplotypes.

The human genome comprises two haploid (i.e. single copy) genomes, one from the father and one from the mother. When we perform conventional genotyping, we usually obtain diploid information, without distinguishing which is from the father and which is from the mother. For example, the “C vs. T” SNP of a person could be one of three diplotypes: CC, CT and TT. A haplotype block is a chromosomal region between two recombination hotspots. Haplotypes are alleles physically residing in a haplotype block of the same haploid chromosome, also called the phase of the genome, as opposed to the “unphased” diplotypes. Since the variants in a haplotype block are assumed to be inherited together, it is possible that a haplotype allele may carry an untyped, disease-causing variant. By analyzing the association between haplotype and the disease trait, we indirectly find the disease-causing variant (Figure 2.2).

A haplotype analysis has many steps:

2.3.1 Capillary Sanger sequencing and next generation sequencing

Conventional capillary sequencers employed the Sanger methodology and served as the major contributors of the human genome project. This method has limitations on sequence reads (~ 1000 bases), which is far shorter than the genome of most species. Thus, sequencing of genomic DNA needs to start by breaking the DNA macromolecule into smaller pieces. Sequencing is then performed on the pieces. The final step is to computationally or manually stitch the pieces together into a complete genome, an inverse problem called assembly. NGS platforms, so-called because of their relative advancement of conventional capillary sequencers in terms of higher speed and lower cost, have been made commercially available. Major vendors (and platforms) include Illumina (Solexa), Roche (454) and Applied Biosystems (SOlid).

However, NGS produces shorter sequence readings than conventional capillary sequencing. The strength of NGS (speed and lower cost) enables a mass production of readings, resulting in deep sequencing with high coverage (i.e. 100x), which may compensate for the weakness of shorter readings. It has been shown that the sequence readings of commercial NGS platforms can be used for the de novo assembly of human genomes (Butler et al., 2008; Gnerre et al., 2011). NGS can also be used to detect germline and somatic point mutations, copy number alterations and structural variations. With the launch of NGS, large-scale genomics investigations now employ NGS in the interest of speed and cost. Conventional capillary sequencing is reserved for small-scale studies. Sequencing technology is still advancing so that new technology (i.e. nanopore) is continuously being developed, so as to reduce the cost and increase the efficiency of sequencing even further.

2.3.2 From sequence fragments to a genome

2.3.3 Signal pre-processing and base calling on SNP arrays

High-density SNP arrays are the major tool of assessing simultaneously the genome-wide variants and CNVs. Based on intensity profiles, the arrays can be used to detect germline and somatic CNVs, where the unusual copy number (other than the two paternal and maternal copies) is shown as increased or decreased intensity signals. In addition to SNP arrays, array CGH is another major platform to detect the CNVs, particularly the loss of heterozygosity in cancer tissues.

Affymetrix and Illumina are two leading commercial manufacturers and vendors of high density SNP arrays. The latest Affymetrix SNP 6.0 microarray product has 1.8 million probe sets targeting 1.8 million loci evenly scattered in the genome (Table 2.1).

Among them, 900,000 are designed to assay SNPs and 900,000 for CNVs. When the number of SNPs is large (e.g. > 100), the high- density microarrays have many practical advantages over various conventional single SNP assays. The latter requires a special set of primers for each SNP. This is not only costly and time-consuming, but also complicates the whole experimental process, including the logistics of reagents and management of data. A good bioinformatics system is essential to keep track of data. Finally, unnecessary variability may be introduced when each SNP is handled at different times and by different persons. A high-density microarray can assay thousands of SNPs in one go, reducing the complexity, cost, time and variability significantly.

High-density SNP microarrays contain pairs of probes designed to capture pairs of DNA sequences harboring the major and the minor allele types respectively (i.e. the target sequence). Existence of a particular genotype in the sample will be detected by the signal of hybridization of the probe corresponding to the target sequence. Hybridization will be shown as intensity signals, which are captured by a scanner and stored as digital images. Raw images usually need to be pre-processed, using background subtraction and normalization techniques (for details, see Section 3.3 in Chapter 3), so as to compensate between array variations. The quantitative intensity signal will then be converted to binary showing exist (1) or non-exist (0) of the allele type in the sample, that is, the base calling process. This is a critical process for reading out nucleotide bases from analog signals.

Many unsupervised and supervised algorithms have been proposed for base calling, such as DM, BRLMM, BRLMM2, Birdseed, Chiamo, CRLMM, CRLMM2 and Illuminus. Since the most recent platforms are capable of detecting structural variations together with SNPs, software such as Birdseed can analyze SNP and CNV at the same time.

The handling of sex chromosome (X and Y) data requires special care compared with autosomal chromosomes. Females have two X chromosomes while males have an X and a Y chromosome. Hence, males only have one allele on X (i.e. hemizygous) and the other “missing” allele can be considered as a null allele (Carlson et al., 2006). Since the null allele may affect the base calling performance when a batch of samples contain both males and females, special algorithms have been developed to handle such occurrences (Carlson et al., 2006). Null alleles may also occur in deletion regions of autosomal chromosomes (Carlson et al., 2006).

2.4.1 Successful factors

Association studies are frequently used to find genes responsible for various clinical manifestations, such as the onset and progression of disease, as well as therapeutic efficacy and toxicity. Association studies were once carried out in a candidate gene fashion, yet now are predominantly done by GWAS. GWAS has successfully identified dozens of novel associations between common SNPs and various common diseases (Hirschhorn, 2009). The knowledge gleaned from these studies is invaluable for the further advancement of biomedical science. Despite some initial success, it was found that most risk alleles have moderate effects and penetrance, preventing them from direct use in personalized healthcare. One of the lessons learned from previous studies is that a well designed study is very important for achieving results. Factors for successful association studies include:

The purpose of all the above factors is to ensure that the detected associations truly represent the clinical trait. A clear and strong contrast in phenotype (i.e. the supercase vs. supercontrol design) could enhance the contrast in genotype, thereby enhancing the possibility of success of finding potential biomarkers, particularly at the exploratory stage. We need to note that the odds ratio (OR) detected does not reflect the true OR in the epidemiological scenario. The hidden population heterogeneity can generally be detected by the inflation factor (Freedman et al., 2004). In such cases, population filters (discussed below) are important.

Many meta-analyses of GWAS have shown that a larger sample size can improve results in terms of both the number of detected SNPs and their significance. For example, a meta-analysis of 46 GWAS with a total of more than 100,000 subjects captured all the 36 originally reported associations, while discovering 59 new associations to personal blood lipid levels (Teslovich et al., 2010). It was found, however, that SNPs on major disease genes of familial hypercholesterolemia were also detected, blurring the conventional categorization of familial diseases vs. common diseases.

2.4.2 The data matrix and quality filters

Association studies are widely used to find the hidden relationship between genetic variants and clinical traits of interest (WTCCC, 2007). In practice, an association study is designed to compare the allele frequencies of two groups with distinct clinical status: the case group comprising subjects with the clinical trait of interest (e.g. a disease state or a therapeutic response) and the corresponding control group. The study subjects are often collected retrospectively, but can also be recruited prospectively. A prospective study means that the clinical outcome is manifested after the samples are collected and assayed (Patterson et al., 2010). The underlying concept for an association study is that genotypes which cause a clinical trait, directly or indirectly, will be more enriched in the case groups, producing a difference in allele frequencies.

In the past, association studies were done by the candidate-gene approach that scrutinizes a targeted set of genes, which is hypothesized to exert effects on disease etiologies based on prior knowledge. This approach has gradually been replaced by GWAS, which is done in a holistic perspective and hypothesis free style with all known genes screened in an unbiased fashion. This was enabled only recently by the maturation of commercial high density SNP assaying platforms such as Affymetrix SNP 6.0 arrays. These platforms were designed for assaying SNPs with not-so-rare minor alleles (e.g. MAF > 0.1), a consequence of the “complex diseases common variant” hypothesis widely accepted before 2009 (Goldstein, 2009). A GWAS usually screens 100,000 to 1,000,000 SNPs, a reasonable sub-sampling of the entire set of genomic variants (Table 2.1).

A conventional master file of an association study is a subject-by-SNP data matrix, comprising genotypes in multiple SNPs of multiple subjects. The genotypes of all subjects are converted from the scanned intensity of SNP microarrays or from the sequencing trace files using the corresponding base calling software. The subjects are sorted by clinical status for ease of comparison. The SNPs are preferably sorted by the chromosomal locations. The subject-by-SNP data matrix is denoted as D, where the subjects are categorized as cases or controls. Each row of D represents a sample of a subject (person), and each column represents an SNP. Denote n as the total number of SNPs, therefore, D = {SNPj | 0 <= j < n}. The subject-by-SNP matrix is part of a PLINK format. It can be visualized using colors, as in the SeattleSNPs presentation. In some data formats. such as the Oxford format, FastTagger and Slide, the SNP-by- subject is used instead. Considering the 100,000 to 1,000,000 SNPs screened in a GWAS, the dimension of the dataset appears to be huge. Yet, the underlying correlation between adjacent SNPs is a biological feature known as LD, which can effectively reduce the complexity of the data matrix.

A collection of quality filters is usually applied to the data matrix, prior to or after the statistical analysis, to ensure the results are derived from high-quality data and reflect the genuine biological effects. These filters are usually imposed in the following order:

The data matrix could therefore contain missing data (blanks) due to the limitation of the base calling algorithm (the null calls) or the sample quality (filtered out by the quality filters). A small degree of missing data will be tolerated and will not hamper the analysis. Nevertheless, we can also choose to employ imputation algorithms to fill in the missing data, or un-genotyped SNPs, based on the LD structure of the SNPs. Imputation algorithms are particularly useful on the meta-analysis of GWAS employing different platforms, for example, Affymetrix and Illumina arrays, where the assayed SNPs are not identical and need to be imputed.

2.4.3 Contrasts of genotypes of individual SNPs

SNPs manifesting strong contrasts of allelic or genotypic frequencies between the clinical groups are pursued in an associations study. Two important gauges for every candidate association are:

The first gauge is the significance level, also known as the P-values, of statistical tests. The P-values quantify the probability of rejecting the null hypothesis when there is no association; that is, the probability of a false positive. Hence, the smaller the P-value, the higher the statistical significance. SNPs with P-values smaller than a predefined threshold (commonly set at 0.05 when only one variant is assessed) are confidently associated to a particular trait of interest. The second gauge is the ORs.

Both P-values and ORs are calculated from the counts of different allelic and genotypic forms. The 2 × 2 and 2 × 3 contingency tables are usually employed to present the counts (Tables 2.2 and 2.3).

Association tests include the allelic, genotypic and trend tests, which can be formulated either as a Chi-square test or a Fisher’s exact test. These tests are used to detect three types of risk patterns, the additive, dominant and recessive modes. Allelic tests evaluate the difference of allelic frequencies between the case and control groups. A related trend test aims to examine the increasing or decreasing trends of diseased frequencies (i.e. proportion of subjects who are the case group) in response to the addition of risk alleles. Allelic and trend tests basically evaluate the additive effect of each risk allele, in the sense that genotype 3 of two risk alleles will have twice the effect as genotype 2 with only one risk allele, thus the relationship between alleles and disease frequency is approximately linear. However, genotypic tests evaluate the difference of genotype frequencies between the case and control groups. Genotypic tests have three variants:

Since the human genome is diploid, an unphased genotype (i.e. the diplotype) contains two alleles. Hence, the total allelic count is twice as large as the total genotype counts. This makes the allelic test double the sample size than the genotypic tests.

For example, the mathematical equation for a Chi-square allelic is: the Chi-square value (X²)

where

and the degrees of freedom is 1.

ORs are the ratio of odds with and without a particular allele or genotype. They can be formulated as:

Under the same formula, OR has many meanings, depending on the definition of G and ~ G. An odds ratio could mean:

The corresponding definitions of G and ~ G are

It is thus important to check the definition before interpreting a result from the literature, particularly when multiple research works are compared. For example, the ORs in Tanaka et al. (2009) are the dominance mode, while those in WTCCC (2007) are heterozygote and homozygote ORs.

A similar concept is relative risk (RR), defined as:

2.4.4 Sample size and multiple comparisons

When genome-wide variants are evaluated in parallel, multiple hypotheses testings, one per a variant, are performed. This comes with the cost that we must tighten the stringency on P-values for each hypothesis. This is called the issue of multiple comparisons. In such types of study, we need to control the family-wise type 1 error rate of the entire study. A common way to address the family-wise error rate is to employ Bonforroni corrections. For example, if n SNPs are assayed in a GWAS, and we want to control the family-wise error rate below 0.05, then each SNP will have an individual P-value smaller than 0.05/n before it can be declared statistically significant. For an Affymetrix SNP6.0 microarray with 900,000 genome-wide SNPs, significance level per SNP based on Bonforroni correction is 5*10⁻⁸.

In the Bonforroni correction setting, n represents the number of independent tests. However, due to the LD between adjacent SNPs in the human genome, hypothesis testings on these SNPs are not independent. Therefore, the conventional Bonforroni correction is often too conservative and inadequate. This can be mitigated by two approaches.

First, a permutation test is a non-parametric alternative that can take LD into consideration. It uses empirical permutation to calculate family-wise error rates, therefore requires a heavy computation which needs to be overcome, particularly in GWAS settings. Permory is a good software of permutation tests on GWAS datasets (Pahl and Schafer, 2010).

Second, prior domain knowledge can be used to devise filters and integrators (Ideker et al., 2011) and reduce the number of independent tests. Filters are used to remove poor-quality data and variants unrelated to the study. Integrators can be used to group variants into higher-order entities such as genes or pathways.

The sample size of this study depends on five factors. The first is the expected effect size of associations, often quantified by ORs. The second issue is the MAF. The range of MAF of SNPs is between 0.01 and 0.5. The third factor is the significance level considering the multiple comparison issues. The fourth factor is statistical power, which is usually set at 80%. The fifth factor is the ratio of the two clinical groups to be compared.

2.4.5 Visual presentation and interpretation

A contrast level analysis of GWAS data will derive a series of P-values for hundreds of thousands of individual SNPs scattered in all chromosomes. A Manhattan plot is a scatter plot that offers a useful grand-scale visualization of the GWAS result. The assayed SNPs are sorted by chromosome locations in the x-axis. The y-axis shows the negative logarithm of P-values, and as a result the higher dots represent significant hits. Freeware, such as WGAviewer (Ge et al., 2008) and Goldsurfer2 (Pettersson et al., 2008), as well as commercial software such as Golden Helix, can be used to produce a Manhattan plot (Figure 2.3).

GWAS and its subsequent validation render a short-list of SNPs, which are statistically associated to clinical traits. Albeit intriguing, the molecular mechanisms behind the associations are still vague. Thus, it is essential to pursue further functional exploration of detected associations. The first step is an in silico check of the SNP (variant) location in relation to the genes and the local LD pattern. Genome browsers in the RefSNP, dbSNP or HapMap websites can help users to locate whether the SNP is in the exon, intron or intergenic regions. If the SNP is located in the intron, then the databases H-DBAS or EBI-ASTD can be used to scrutinize the alternative splicing patterns. H-DBAS can also illustrate whether the SNP is in coding or UTR regions. It also offers a nice comparison with the mouse transcripts.

A check of the functional domain where the SNP resides requires a protein-level database, such as Pfam or InterPro (Box 2.1). A complete understanding of the association requires data from further experiments on cell or tissue specific gene expressions.

2.4.6 The missing heritability

The GWAS approach has successfully identified and validated hundreds of disease prone (and occasionally disease bearing) SNPs. These SNPs may serve as clinically useful biomarkers and biosignatures for disease susceptibility and prognosis, as well as revealing disease mechanisms (Hirschorn, 2009). However, this approach also has its limitations. It has been observed that GWAS-detected alleles are of low penetrance and that the total variability explained by all these SNPs together is only a fraction of the variation caused by heritability (Doucleff, 2010; Donnelly, 2008; Goldstein, 2009; McCarthy, 2008, Manolio et al., 2009). It is thus suggested that personal disease risks are not only contributed by common SNP alone, but also by other forms of genomic variations such as CNVs or rare mutations, as well as their synergistic effects (Goldstein, 2009; McClellan and King, 2010; Manolio et al., 2009).

Among these possible sources of missing heritability, the accumulation of multiple mutations (i.e. MAF < 1%) for disease etiology deserves most attention. Lessons from the familial diseases, such as familial hypercholesterolemia, familial breast cancer, hereditary nonpolyposis colorectal cancers and cystic fibrosis, tell us that numerous rare germline mutations play significant roles. From an evolutionary perspective, all SNPs originate from mutations. They are actually two sides of the same coin (McClellan and King, 2010). The importance of mutations justifies the sequencing technology as a critical tool for detecting the genomic etiology of disease.

1. element insertion;

2. element deletion;

3. element substitution;

4. operators ·/+ swap; and

5. case and control swap.

2.7.1 Chromothripsis on cancer progression

One major characteristic in the cancer genome is its pervasive structural variations caused by extensive rearrangements. It has long been hypothesized that the structural variations gradually accumulate over time. Contrary to this model, a mechanism called chromothripsis was recently discovered. It refers to localized massive rearrangements occurring in one or few chromosomes, which may occur in one catastrophic event (Stephens et al., 2011).

Stephens et al. (2011) observed an infrequent, somatic, localized rearrangement pattern of the cancer genome based on their NGS and SNP microarray assays. This pattern has three major features:

They hypothesized that this pattern is caused by rearrangements occurring in a single catastrophic event, rather than a series of events. They employed a Monte Carlo method to simulate both the one-event and the progressive models, and concluded that the one-event model better depicts their observation. Furthermore, this event is shown to occur early, due to the paucity of other mutations occurring alongside the chromothripsis. Thus, this early event, although infrequent, disrupts multiple genes so as to drive the further progression of cancer. This example demonstrates the analysis of the cancer genome to shed light on an obscure yet important molecular event during the progression of the disease.

2.7.2 Ancestry inference and population analysis based on genomic variants

High resolution genomic variants offer rich information about human population and ancestry, in addition to all the health-related messages mentioned above. A few personal genome service companies, such as decodeMe, 23andMe and Navigenics, have emerged since 2005 upon the availability of technology. All of them provide population analysis for individual customers. Conventionally, ancestry inference and population analysis were carried out using mitochondrial DNA or the Y chromosome. Now with the availability of genome-wide and high resolution variant information, a higher resolution on ancestry inference and population analysis can certainly be achieved.

Li and Durbin (2011) demonstrate the use of genome-wide variants for the inference of human ancestry and population history. The analysis is based on the genome of seven subjects. The local density of heterozygosity is the key clue to the estimation of local time.

Behar et al. (2010) provide a population analysis of Jewish people based on genotypes of 121 persons from Illumina 610 K and 660 K bead array. They used principal component analysis (PCA) to project the samples in a 2-D plane spanned by two major principal components, which correspond nicely to two geographical axes. The technique of PCA will be introduced in Chapter 3.

 Reference genomes are constructed by the de novo assembly of sequence reads.

 Sequence alignments are basic skills to find homologous sequences from the public domain database.

 A spectrum of variants (SNP, CNV and structural variations) exist in the human genome.

 Association studies aim to associate the variant genotypes to clinical traits, thereby revealing phenotype critical genes.

 The strength of association between genomic variants and clinical traits is quantified by P-values and ORs.

 Manhattan plots can be used for genome-wide presentation of association.

 Adaptive models are useful for capturing clinical genetic signatures.

 Somatic variants, particular the structure variations, are responsible for cancer initiation and progression.

Altshuler, D.M., Gibbs, R.A., Peltonen, L., et al. Integrating common and rare genetic variation in diverse human populations. Nature. 2010; 467(7311):52–58.

Ashworth, A., Lord, C.J., Reis-Filho, J.S. Genetic interactions in cancer progression and treatment. Cell. 2011; 145(1):30–38.

Behar, D.M., et al. The genome-wide structure of the Jewish people. Nature. 2010; 466:238–242.

Beroukhim, R., Mermel, C.H., Porter, D., Wei, G., Raychaudhuri, S., et al. The landscape of somatic copy-number alteration across human cancers. Nature. 2010; 463:899–905.

Butler, J., MacCallum, I., Kleber, M., et al. ALLPATHS: de novo assembly of whole-genome shotgun microreads. Genome Res. 2008; 18(5):810–820.

Carlson, C.S., et al. Direct detection of null alleles in SNP genotyping data. Hum. Mol. Genet. 2006; 15(12):1931–1937.

Carlson, C.S., Eberle, M.A., Rieder, M.J., Yi, Q., Kruglyak, L. Selecting a maximally informative set of single-nucleotide polymorphisms for association analyses using linkage disequilibrium. Am. J. Hum. Genet. 2004; 74:106–120.

Chen, E.Y., et al. PolyScan: An automatic indel and SNP detection approach to the analysis of human resequencing data. Genome Res. 2007; 17:659–666.

Cordell, H.J. Detecting gene-gene interactions that underlie human disease. Nat. Rev. Genet. 2009; 10:392–404.

DeCerbo, J., Carmichael, G.G. SINEs point to abundant editing in the human genome. Genome Biol. 2005; 6(4):216–219.

Donnelly, P. Progress and challenges in genome-wide association studies in humans. Nature. 2008; 456:728–731.

Doucleff, M. Genomics select. Cell. 2010; 142(2):177.

Ewing, B., Green, P. Base-calling of automated sequencer traces using phred. II: Error probabilities. Genome Res. 1998; 8:186–194.

Ewing, B., Hillier, L., Wendl, M.C., Green, P. Base-calling of automated sequencer traces using phred I: Accuracy assessment. Genome Res. 1998; 8:175–185.

Fan, H.C., Gu, W., Wang, J., et al. Non-invasive prenatal measurement of the fetal genome. Nature. 2012; 487(7407):320–324.

Feuk, L., Carson, A.R., Scherer, S.W. Structural variation in the human genome. Nat. Rev. Genet. 2006; 7:85–97.

Freedman, M.L., Reich, D., Penney, K.L., McDonald, G.J., Mignault, A.A., et al. Assessing the impact of population stratification on genetic association studies. Nat Genet. 2004; 36(4):388–393.

Ge, D., Zhang, D., Need, A.C., Martin, O., Fellay, J., Telenti, A., Goldstein, D.B. WGAViewer: Software for genomic annotation of whole genome association studies. Genome Res. 2008; 18(4):640–643.

Gnerre, S., et al. High-quality draft assemblies of mammalian genomes from massively parallel sequence data. PNAS. 2011; 108:1513–1518.

Goldstein, D.B. Common genetic variation and human traits. N. Engl. J. Med. 2009; 360:1696–1698.

Goodnow, C.C. Multistep pathogenesis of autoimmune disease. Cell. 2007; 130:25–35.

Gupta, G.P., Massague, J. Cancer metastasis: Building a framework. Cell. 2006; 127:679–695.

Hanahan, D., Weinberg, R.A. Hallmarks of cancer: The next generation. Cell. 2011; 144:646–674.

Hirschhorn, J.N. Genome-wide association studies: Illuminating biologic pathways. N. Engl. J. Med. 2009; 260:1699–1701.

Hsieh, C.H., Liang, K.H., Hung, Y.-J., Huang, L.-C., Pei, D., et al. Analysis of epistasis for diabetic nephropathy among Type 2 diabetic patients. Hum. Mole. Gen. 2006; 15(18):2701–2702.

Ideker, T., Dutkowski, J., Leroy Hood, L. Boosting signal-to-noise in complex biology: Prior knowledge is power. Cell. 2011; 144:860–863.

IHGSC. Finishing the euchromatic sequence of the human genome. Nature. 2004; 431:931–945.

Kitzman, J.O., Snyder, M.W., Ventura, M., et al. Non-invasive whole-genome sequencing of a human fetus. Sci Transl Med. 2012; 4(137):137–176.

Lander, E.S., Linton, L.M., Birren, B., et al. Initial sequencing and analysis of the human genome. Nature. 2001; 409:860–921.

Langmead, B., Salzberg, S. Fast gapped-read alignment with Bowtie 2. Nature Meth. 2012; 9:357–359.

Langmead, B., Trapnell, C., Pop, M., Salzberg, S.L. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 2009; 10:R25.

Li, H., Durbin, R. Inference of human population history from individual whole-genome sequences. Nature. 2011; 475(7357):493–496.

Liang, K.H., Wu, Y.-J. Prediction of complex traits based on the epistasis of multiple haplotypes. J. Hum. Gen. 2007; 52(5):456–463.

Liang, K.H., Hwang, Y., Shao, W.-C., Chen, W.Y. An algorithm for model construction and its applications to pharmacogenomic studies. J. Hum. Gen. 2006; 51:751–759.

Lin, L., Shen, S., Tye, A., et al. Diverse splicing patterns of exonized Alu elements in human tissues. PLoS Genet. 2008; 4:e1000225.

Liu, G., Wang, Y., Wong, L. FastTagger: An efficient algorithm for genome- wide tag SNP selection using multi-marker linkage disequilibrium. BMC Bioinformatics. 2010; 11:66.

Luebeck, E.G. Genomic evolution of metastasis. Nature. 2010; 467:1053–1054.

Manolio, T.A., et al. Finding the missing heritability of complex diseases. Nature. 2009; 461:747–753.

McCarthy, M.I. Casting a wider net for diabetes susceptibility genes. Nat. Genet. 2008; 40:1039–1040.

McClellan, J., King, M.-C. Genetic heterogeneity in human disease. Cell. 2010; 141(2):210–217.

Minelli, C., Thompson, J.R., Abrams, K.R., et al. How should we use information about HWE in the meta-analysis of genetic association studies? Int. J. Epidemiol. 2008; 37:136–146.

Muotri, A.R., et al. The necessary junk: New functions for transposable elements. Hum. Mol. Genet. 2007; 16:R159–R167.

Nacheva, E.P., Brazma, D., Virgili, A. Deletions of immunoglobulin heavy chain and T cell receptor gene regions are uniquely associated with lymphoid blast transformation of chronic myeloid leukemia. BMC Genomics. 2010; 11:41.

Noe, L., Kucherov, G. YASS: Enhancing the sensitivity of DNA similarity search. Nucleic Acids Res. 2005; 33(2):W540–W543.

Pahl, R., Schafer, H. PERMORY: an LD-exploiting permutation test algorithm for powerful genome-wide association testing. Bioinformatics. 2010; 26(17):2093–2100.

Patterson, S.D., et al. Prospective-retrospective biomarker analysis for regulatory consideration: White paper from the industry pharmacogenomics working group. Pharmacogenomics. 2010; 12(7):939–951.

Pettersson, F., Morris, A.P., Barnes, M.R., Cardon, L.R. Goldsurfer2 (Gs2): A comprehensive tool for the analysis and visualization of genome wide association studies. BMC Bioinformatics. 2008; 9:138.

Pennisi, E. 1000 Genomes project gives new map of genetic diversity. Science. 2010; 330:574–575.

Phillips, P.C. Epistasis: The essential role of gene interactions in the structure and evolution of genetic systems. Nat. Rev. Genet. 2008; 9(11):855–867.

Pruim, R.J., Welch, R.P., Sanna, S., et al. LocusZoom: Regional visualization of genome-wide association scan results. Bioinformatics. 2010; 26(18):2336–2337.

Pruitt, K.D., Tatusova, T., Brown, G.R., Maglott, D.R. NCBI Reference Sequences (RefSeq): Current status, new features and genome annotation policy. Nucleic Acids Res. 2012; 40:D130–D135. [(Database issue)].

Stephens, P.J., et al. Massive genomic rearrangement acquired in a single catastrophic event during cancer development. Cell. 2011; 144:27–40.

Tanaka, Y., Nishida, N., Sugiyama, M., et al. Genome-wide association of IL28B with response to pegylated interferon- and ribavirin therapy for chronic hepatitis C. Nature Genet. 2009; 41(10):1105–1109.

Teslovich, T.M., Musunuru, K., Smith, A.V., et al. Biological, clinical and population relevance of 95 loci for blood lipids. Nature. 2010; 466(7307):707–713.

The International HapMap Consortium. A second generation human haplotype map of over 3.1 million SNPS. Nature. 2007; 449:851–862.

Tregouet, D.A., Garelle, V. A new JAVA interface implementation of THESIAS: Testing haplotype effects in association studies. Bioinformatics. 2007; 23(8):1038–1039.

Wang, K., et al. PennCNV: An integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data. Genome Res. 2007; 17:1665–1674.

Wecks, S., Del-Favero, J., Rademakers, R. novoSNP, a novel computational tool for sequence variation discovery. Genome Res. 2005; 15(3):436–442.

WTCCC. Genome-wide association study of 14,000 cases of 7 common diseases and 3000 shared controls. Nature. 2007; 447:661–678.

Yachida, S., Jones, S., Bozic, I., et al. Distant metastasis occurs late during the genetic evolution of pancreatic cancer. Nature. 2010; 467(7319):1114–1117.

Zhang, J., et al, SNPdetector: A software tool for sensitive and accurate SNP detection. PLoS Computational Biology. 2005. [October].