3.1.1 Big Data

The term big data is applied to large volumes of biological data, as it is to data derived from financial or astronomical domains. Each area generates petabytes of data on a regular basis. As in all other areas of big data endeavour, the computational techniques appropriate to the size of these data streams are appropriate to apply. These include artificial intelligence techniques of linear and non‐linear regression, classification and machine learning.

3.1.2 Computational Challenges

There are significant challenges in analysing sequence data for genomics. These include:

• Computer hardware utilisation, for example, exploitation of graphics chips (GPUs) designed for rapid matrix calculations (as demanded by machine learning)
• Data storage, covering parallelised file systems, object stores and hierarchical storage
• Software development, necessitating the expertise in writing code for making effective and efficient use of hardware and ever increasing volumes of data.

Cloud‐based approaches are coming to the fore, with commercial organisations offering viable solutions and research institutions implementing internal cloud‐based flexible compute environments. This approach enables a timely response to the rapidly changing flow of data and the information and knowledge derived from it. In terms of software advances, programming for machine learning and artificial intelligence has much to contribute in identifying patterns and trends when analysing large quantities of genomic data. Comparison of whole genome sequences and variant call files (VCFs) resulting from these large sequencing projects are areas of active research [3].

3.1.3 Bioinformatics Roles

Bioinformatics scientists who assist in the design of experiments for research programmes both large and small rely on the work of the primary resource annotators. These annotators are proficient at reviewing the literature related to the part of the genome they are annotating, and running and interpreting the results of sequence alignment software. A second group of bioinformaticians, who are skilled at using the many tools of bioinformatics, assist in the selection and generation of oligonucleotides (e.g. polymerase chain reaction [PCR] or sequencing primers) for generating genetic engineering designs. These designs are key to the execution of experimental work in the laboratory. They use the annotation of genomic sequences by focusing on individual genes to support experimental design.

As the scope of the research programmes becomes more ambitious, higher throughput systems are put in place, including laboratory automation and high throughput techniques based on multiwell plates (e.g. 96‐well and 384‐well plates). The amount of material generated by these high throughput processes demands ever higher capacity in sequencing capability to confirm what has happened in these experiments (known as genotyping). Thus there are additional bioinformatics roles involved in putting together pipelines of bioinformatics tools (aligners, quantifiers, visualisations, further database storage of results, laboratory information management systems [LIMSs], reporting systems and so on).

Additional roles encompass the analysis involved in the related areas of transcriptomics (the study of the complete transcript set from cells) and epigenomics (the study of all the epigenetic modifications to the cell's genetic material, including methylation of DNA and histone modification). These roles may involve assessing which genes in a whole genome are affected when a cell is subjected to some stress, like a drug or heat or light. Again, annotation is key to accurately map the transcripts back to the correct gene or non‐coding RNA. Epigenomics and transcriptomics data also drive a more accurate annotation of the genome.

Therefore, some bioinformatics scientists will be intimately involved in experimental design and, indeed, many laboratory‐based scientists will have significant bioinformatics skills in their area of expertise. Even now, most experiments contain at least some bioinformatics analysis. The sheer bulk of data means that it is often computational analysis that leads the experimental hypotheses, or at least allows for more rapid removal of hypotheses that are wrong. Other bioinformatics scientists will be much farther removed from the molecular biology in the laboratory. Some will need advanced computational and software development skills; others may be able to rely on publicly available or commercial bioinformatics tools to support them in their experimental design or analytical work.

Different business needs will also define the role of the bioinformatics scientist. In a pharmaceutical preclinical research team, bioinformatics may be used to understand the evolutionary relationships between members of a family of genes. This helps determine whether a specific gene is being targeted by a chemical compound (e.g. a drug) or whether the compound may affect an unintended but closely related member of the protein family. Bioinformatics may be used in the following ways (among others):

• Curate in‐house databases of drug targets (generally proteins, which are normally enzymes catalysing key reactions).
• Select antigenic sites for potential antibody‐based therapies.
• Design gene editing protocols in a precision genome engineering context to support a drug discovery programme.

Some research contexts may require extensive knowledge of structural biology and protein DNA complexes as well as complexes involving RNA, other proteins or small molecules. This area pushes more into the disciplines of biophysics and computational chemistry, which are both close scientific neighbours of bioinformatics. Images and the metadata that are assigned to them by technicians and clinicians are important in the context of clinical bioinformatics.

Whatever the context of the bioinformatics, there are some fundamental theoretical concepts that should be understood by all scientists as being characteristic of bioinformatics as a science.

3.2.1 Molecular Biology

Whatever the type of bioinformatics role and whatever the scale of data to be analysed, all must understand the fundamental molecular biology that underpins the biology and the generation of biological data. Core to this understanding is the central dogma of molecular biology: ‘DNA makes RNA makes protein’ (a useful, if oversimplistic, memory phrase), but more formally ‘DNA is transcribed to produce RNA which is translated to produce protein’. Fundamentally, the central dogma is about the flow of information from DNA, via RNA to the protein product (as stated by Francis Crick in 1956).

From the point of view of bioinformatics, there is a one‐way flow of information from DNA to DNA, from DNA to RNA and from RNA to protein. There is no flow of information from protein back to RNA or DNA. Proteins can and do interact with DNA and RNA but there is no information transfer. The information we are describing here is the sequence of DNA, segments of which are transcribed to produce an RNA template. This is then translated three bases (a codon) at a time into the amino acids that form the single polypeptide chain of a protein. Knowing the sequence of bases in a stretch of DNA (or RNA) allows us to predict the resulting protein sequence. Knowing a protein sequence does not allow us to predict with any certainty the sequence of bases that generated that protein. This is because the genetic code is a redundant code with multiple codons for the same amino acid residue. There are only two exceptions: there is only one codon for methionine (M), which has a special role as the first residue in a nascent polypeptide sequence in RNA translation, and only one codon for tryptophan (W). These two amino acids are generally the least commonly observed residues, although, in individual protein sequences, biophysical requirements may mean that other residues have a lower relative abundance. In fact, there is a large degree of variability in the total abundance of amino acids in a protein. Cysteine is a specific example of an amino acid that is more commonly observed in the shorter sequences of the polypeptide hormones. Here, the strong covalent bonding of the disulphide bridge is necessary to stabilise the structure of the smaller molecule in an extracellular environment. This is one of the major reasons for the existence of disulphides. In genomic bioinformatics, we are always mindful that the workhorses of the cell are proteins.

You should use reliable resources from the reference list at the end of this chapter to review the following topics:

• The structure of DNA
• The four DNA bases (A, T, C, G)
• Complementary base‐pairing in DNA
• The four bases of RNA (A, U, C, G) and their base‐pairing
• The production of messenger RNA (mRNA) from the DNA template in the genome
• The production of the primary protein sequence (consisting of the 20 naturally occurring amino acids: A, C, D, E, F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W, Y)
• The physical characteristics of the amino acid residues (see Table 3.1).

Note that generally in bioinformatics the one‐letter codes for the amino acid residues are used. See Table 3.1 for the one letter, three letter and full names grouped according to physical characteristics along with their codons.

Fundamental to the skill of bioinformatics is the ability to handle sequence data either for an individual gene or part of a gene and its protein product. Therefore, we need to be able to convert information at DNA or RNA levels to protein sequence efficiently. We can then work on multiple genes, or indeed whole genomes, effectively. This process is known as three‐frame translation (or six‐frame if both strand directions are taken into account). The longest open reading frame (ORF) – that is, the longest sequence of residues translated without encountering a stop codon – is generally taken to be the standard protein product of the nucleic acid sequence. See Figure 3.1 for an illustration of the translation tracks turned on in the Ensembl genome browser and Figure 3.2 for a closeup zoomed in detail in which the translations in all six reading frames are made clear.

3.2.2 Gene Structure

Different genomes have different levels of complexity. The human genome is perhaps surprisingly not the most complex genome we know. It has a relatively low complement of genes (around 23 000) compared to some other organisms. It is the complex structure of human genes that makes for a combinatorial explosion of gene and protein products. The processes involved result from regulation of gene expression and splicing of exons and introns. The memory phrases ‘exons are expressed’ and ‘introns interrupt’ can be helpful. Introns are eventually spliced out by the editing machinery of the cell. Many human genes, for example, have multiple splice variants that are made up by editing together specific exons of one particular gene. However, the most complex example of splice variation known is the gene DSCAM in Drosophila melanogaster. With 38 000 splice forms, this single gene has more splice variants than there are genes in the entire D. melanogaster genome. This can make analysis a complex process. Not all splice variants are known and many are tissue, or even cell specific – that is, they are only expressed in certain cells and at certain stages of the cell cycle (or upon cell–cell signalling or stresses). When conducting an analysis, we may refer to the canonical sequence. This is a sequence that reflects the most commonly observed base at a position.

3.2.3 Software Development Note

Bioinformatics scientists involved in programming applications for analysis should note that application programming interfaces (APIs) are available to assist in gathering the data required for conducting a bioinformatics analysis. These can be used in scripts and programs written in scripting languages like Perl and Python. For example, in the ENSEMBL core API, the canonical sequence refers to a commonly observed arrangement of exons in a particular transcript. Other transcripts may be observed, or might be predicted to occur based on evidence of splicing observed in sequencing projects. The API calls can be used to determine which transcript is most appropriate to use. Discussion with domain specialist scientists may also be necessary at this point in a project.

3.3.1 Accessing and Using Data

Data are made available using the Internet and are typically accessed using various web browsers from many providers. Much of the data used in bioinformatics is available publicly and freely via file download sites. These are typically FTP servers, but as file sizes get ever larger alternative methods of distribution are sometimes preferred. Cloud‐based services from a number of vendors are available that provide a flexible compute capability alongside databases. Data can thus be analysed in one place without having to replicate it across multiple sites. There are, of course, costs associated with computing in the cloud as there are in bringing data in‐house and providing computational facilities to analyse it.

You should be familiar with at least one web‐based genome browser. The browsers at www.ensembl.org and genome.ucsc.edu are good places to start exploring and we use the Ensembl genome browser to illustrate several techniques in the following section.

3.4.1 Genome Browsers

Here we will use the www.ensembl.org site, which provides a fully capable genome browser enabling users to explore the publicly available data for specific genomes. Often bioinformaticians will work on a single gene or family of genes. Scientific colleagues working on a project will already have the specific code to look up for a gene. As a bioinformatician, you will want to become familiar with the background to the project, which suggests a wider search. Assuming that the project is researching breast cancer genes, we would probably first use a trusted web search engine to get a general impression of the work going on in the general area of ‘breast cancer genes’. This may seem obvious and unnecessary but the effort that large companies put into development of search engines can shortcut a lot of hard bioinformatics work as new resources are coming online all the time.

Next, use a tool like the Open Targets Platform (which is freely accessible at www.targetvalidation.org) to perform a more targeted search resulting in the output shown in Figure 3.7.

The first thing to be aware of is that gene symbols change over time. What we currently call BRCA1 has synonyms in the scientific literature. These synonyms are listed in the Ensembl page for the human BRCA1 gene. The Ensembl page (a portion of which is shown in Figure 3.8 and a part of the scrollable interactive view in Figure 3.9) is a very functional and beautiful representation of a great many pieces of data brought together to provide a rich environment for genomic exploration. Note the number of transcripts and the link to splice variants. This is a complicated gene, which has been observed in 30 different isoforms. There are 97 versions of this gene in other organisms. These are known as orthologues – genes that have evolved from a common ancestral gene over time as populations also evolved to become distinct species (termed speciation).

3.4.2 Sequence Comparison

We compare sequences (whether DNA, cDNA, RNA or protein) on a regular basis. We do this to discover common regions, understand overall structures, predict functionality of unknown regions from similarity with known ones, understand evolutionary relationships, etc. Sequence alignments using sequences of a similar length and showing a high degree of sequence similarity are easier to interpret. Given a sequence to investigate, we would run a database search (normally using one of the Basic Local Alignment Search Tool (BLAST) suite of programs, viz. www.ensembl.org/Homo_sapiens/Tools/Blast) to find related sequences. Comparing a shorter sequence (perhaps of a few hundred bases) with a much longer sequence – say a chromosome millions of bases long – is a different task and uses specially configured database search tools.

Whichever tools we use, the concepts of local and global alignment (see Section 3.2.3) become important. In genomic bioinformatics, we find ourselves comparing DNA with DNA much of the time. In other areas of bioinformatics, we may use protein/protein comparisons more. The basic rules for comparing sequences are the same in either case, but the alphabet for proteins is more extensive than for DNA. DNA comparison programs tend to be optimised for finding short stretches of identical sequence and then stringing these regions together. Protein level comparisons will normally use tables of similarity scores (‘scoring matrices’) to assess regions of similarity and scoring priority.

3.4.2.2 The Dotplot

A useful visualisation tool is the dotplot. Here, two sequences (either DNA or protein) are laid out one on the X‐axis and the other on the Y‐axis of Cartesian space. We end up with a rectangle of cells that are marked if the XY positions have identical bases/residues. Thus we assess each cell in turn and visualise the result. More sophisticated algorithms introduce a sliding window, with a number of positions contributing to the score at any particular cell. A self/self sequence comparison is shown in Figure 3.10. Dotplots are very useful when reviewing sequence data at the genomic level following WGS as they clearly indicate structural variants (see Figure 3.11).

3.5.1 CRISPR/Cas9 and Off‐Targets

Cas9 is an enzyme (a protein) that can induce a double‐stranded break in DNA. It was discovered in the Streptococcus pyogenes bacterium where it assists the bacterium to remember that it has seen an invading DNA virus before. It does this by incorporating the invading viral DNA into its own genome and by using sequence complementarity to recognise and cut the DNA of the virus should it infect the bacterium again. It was subsequently discovered that the Cas9 protein was responsible for this sequence‐specific recognition and that this could be reprogrammed to recognise desired genomic target sites. Importantly, it could also be transferred to work in mammalian cells to cut double‐stranded DNA at a defined location. The era of precision genomics was moved on significantly.

The repair of the double‐strand break by the cellular machinery is most frequently achieved by the non‐homologous end joining (NHEJ) process. This process is somewhat error prone and leads to short insertion or deletion events at the cut site. Less frequently, a mechanism known as homology directed repair (HDR) is used by the cell. This relies on copying the correct sequence from a template DNA (naturally the sister chromatid). This can be exploited to introduce defined changes and insertions of longer stretches of DNA into the genome. The results of experiments directed at specific genes use bioinformatics techniques to design primers and to prioritise CRISPR sites. The experiment can be evaluated by NGS genotyping to find out what the effect of the DNA repair has been.

The Cas9 protein scans the genome and stutters over pairs of G bases. The protein can be programmed to latch on to a specific genomic sequence of around 20 bases ending in ‘NGG’ (where ‘N’ is any base). The programming tool is a 20‐base guide RNA that is provided along with the Cas9 protein. There are about 300 million ‘NGG’ sites in the human genome. Furthermore, there can be off‐target cutting effects where there are one or more mismatches between the RNA guide and its target.

Figure 3.12 illustrates the genome browser interface of the WGE Crispr design tool (www.sanger.ac.uk/htgt/wge). This tool is a bioinformatics tool that assists in the selection of sites for CRISPR/Cas9 experiments. It supports many different features, including interactive searching for CRISPR/Cas9 sites in human and mouse genomes.

The WGE implementation consists of three logical parts but users are normally only aware of the web application. This provides the context for searching and the visualisation of the search results. WGE also links to (i) a relational database (actually, PostgreSQL) that provides a persistent store of CRISPR and off‐target information and (ii) a CRISPR‐Analyser tool that runs on a separate server that provides fast indexed lookups to all the off‐targets in the human and mouse genomes with up to 4 mismatches to the 20‐base guide RNA query. The CRISPR‐Analyser provides its results principally in the form of an off‐target string. The off‐target string shown in Figure 3.12 is {0: 1, 1: 0, 2: 2: 1, 3: 7, 4: 89}. This means that there is one sequence in the human genome that matches the CRISPR guide exactly (this is the query or on‐target sequence). There are no hits with just 1 mismatch in the guide sequence; 1 hit with 2 mismatches, 7 hits with 3 mismatches and 89 hits with 4 mismatches. The off‐target string is a helpful data structure that is used to help prioritise CRISPR guides for experimental design [11].

For the bioinformaticians, it is crucial to understand these basic mechanisms and to come up with computational solutions – either by programming new systems or by using tools written by others that can form part of a bioinformatics pipeline.

In researching, designing and implementing the WGE system many decisions had to be made about what exactly would constitute a ‘Crispr’ object in the system. Considerations of space for storage, rapidity of access to on‐target and off‐target data and summarisation of information all had to be carefully balanced to create a responsive system. Even the fine details of assigning IDs to Crisprs became critical (i.e. they cannot change once assigned) when a feed of Crispr locations was provided to the Ensembl genome browser.

The source code for WGE and the CRISPR‐Analyser is available on GitHub at github.com/htgt/WGE and github.com/htgt/CRISPR‐Analyser. The WGE code is written mainly in Perl and Javascript, and the CRISPR‐Analyser in C++.

3.5.2 Drug Discovery and Target Discovery

Bioinformatics has contributed significantly to drug discovery (discovery of new medicines) since the early 1990s when sequencing programs in pharmaceutical companies were introduced to trawl short‐read sequences, called expressed sequence tags (ESTs), for new variants of known genes. Drug targets are typically proteins that are amenable to having their function modulated by small molecule ligands. Such ligands are known as drugs (medicines). Now that we have more advanced sequencing capabilities, transcriptomics has become a major research endeavour. Projects underway are aimed at sequencing tissues (e.g. GTeX consortium) and single cells (e.g. HCA – Human Cell Atlas) from a wide variety of sources and mapping their transcripts to build up a profile of the genes that are expressed in individual cells across whole organisms.

The Human Cell Atlas (www.humancellatlas.org) is a major international collaboration in this area whose mission is ‘to create comprehensive reference maps of all human cells – the fundamental units of life – as a basis for both understanding human health and diagnosing, monitoring, and treating disease’.

The GTeX Genotype Tissue Expression consortium (www.gtexportal.org) catalogues genetic variation and the influence of that variation on gene expression within and between all major tissues in the human body. By building up a database and associated tissue bank of all major tissues from 1000 individuals, bioinformaticians have a rich resource of genetic variation, gene expression, histological and clinical data to mine.

Bioinformatics techniques are intimately involved in discovering new drugs and repurposing known drugs for new uses. Bioinformatics has been used to help understand the mechanism of drug action and through phylogenetics and the elucidation of evolutionary pathways has been able to clarify disputed molecular mechanisms [12]. Often highly divergent species are involved in the analysis and new approaches for improving phylogenetic methods are applied in the context of phylogenomics [13] – i.e. phylogenetics applied at the genomic scale.

3.5.3 Transcriptomics

When we are interested in looking at the expression of the complement of genes in a cell, we need to sequence the transcripts that have been transcribed from the genomic DNA into messenger RNA. In drug discovery, understanding the genes that are actually transcribed (and later expressed) is critical to discovery of new therapeutic targets. Generally, small molecule drugs will be targeted at specific proteins, rather than DNA or RNA. Discovering all the proteins expressed by a cell is done using biophysical techniques, including advanced mass spectrometry and the tools of proteomics. It can be effective simply to look at mRNA transcripts, a technique that also provides the ability to count transcripts and to see the difference in transcription levels between cells in healthy and diseased states, or at different stages in development.

The current RNA‐Seq technology that can be used to analyse RNA transcripts relies on transcribing the RNA transcripts back to DNA for sequencing, so it is in fact a DNA sequencing technique. This involves both copying (reverse transcription) and amplification (PCR) steps. From a bioinformatics perspective this creates problems because we have to assume that the reverse transcription and PCR processes have amplified the transcripts uniformly in generating the cDNA library, which may not always be the case. Transcript counting depends upon this assumption.

Newer native RNA‐Seq technologies (for example, Oxford Nanopore Technologies) remove the necessity for reverse transcription to cDNA and enable single longer molecules of RNA to be sequenced. This has the added advantage of sequencing through splice sites and making alternatively spliced transcripts easier to count [14]. This ability to derive reliable transcript counts is important because experiments often involve comparing gene expression between different conditions, such as drug treated compared to non‐drug treated.

Using these and associated genomics strategies, there is great potential to improve drug targeting and creation of diagnostic tools with increased sensitivity. The role of bioinformatics is central in optimising algorithms for alignment, quality metrics, standards for variant calling and interpretation as well as making the results available globally. This will ultimately foster a deeper understanding of disease and lead to greater therapeutic precision [15].

References

1. 1 International Genome Sequencing Consortium (2001). Initial sequencing and analysis of the human genome. Nature 409: 860–921.
2. 2 Ledford, H. (2016). AstraZeneca launches project to sequence 2 million genomes. Nature 532: 427.
3. 3 Zook, J., Chapman, B., Wang, J. et al. (2014). Integrating human sequence data sets provides a resource of benchmark SNP and indel genotype calls. Nat. Biotechnol. 32: 246–251.
4. 4 Livingstone, C.D. and Barton, G.J. (1993). Protein sequence alignments: a strategy for the hierarchical analysis of residue conservation. Comput. Appl. Bio. Sci. 9: 745–756.
5. 5 Attwood, T.K. and Parry‐Smith, D.J. (1998). Introduction to Bioinformatics. London: Pearson.
6. 6 Sievers, F., Wilm, A., Dineen, D.G. et al. (2011). Fast, scalable generation of high‐quality protein multiple sequence alignments using Clustal Omega. Mol. Syst. Biol. 7: 539.
7. 7 Parry‐Smith, D.J., Payne, A.W.R., Michie, A.D., and Attwood, T.K. (1997). CINEMA ‐ a novel Colour INteractive Editor for Multiple Alignments. Gene 211 (2): GC45–GC56.
8. 8 Finn, R.D., Attwood, T.K., Babbitt, P.C. et al. (2016). InterPro in 2017 – beyond protein family and domain annotations. Nucleic Acids Res. 45 (D1): D190–D199.
9. 9 Li, H., Handsaker, B., Wysoker, A. et al. and 1000 Genome Project Data Processing Subgroup (2009). The sequence alignment/map (SAM) format and SAMtools. Bioinformatics 25: 2078–2079.
10. 10 Abuín, J.M., Pichel, J.C., Pena, T.F., and Amigo, J. (2015). BigBWA: approaching the burrows–wheeler aligner to big data technologies. Bioinformatics 31 (24): 4003–4005.
11. 11 Hodgkins, A., Farne, A., Perera, S. et al. (2015). WGE: a CRISPR database for genome engineering. Bioinformatics 31 (18): 3078–3080.
12. 12 Xuhua, X. (2017). Bioinformatics and drug discovery. Curr. Top. Med. Chem. 17 (15): 1709–1726.
13. 13 Parry‐Smith, D.J. (2003). Bioinformatics: its role in drug discovery. In: Burger's Medicinal Chemistry and Drug Discovery, 6e, vol. 1 Drug Discovery (ed. D. Abraham). Hoboken, New Jersey, USA: Wiley.
14. 14 Hussain, S. (2018). Native RNA‐sequencing throws its hat into the transcriptomics ring. Trends Biochem. Sci. 43 (4): 225–227.
15. 15 Ashley, E.A. (2016). Towards precision medicine. Nat. Rev. Genet. 17 (9): 507–522.

Further Reading

1. Lodish, H. et al. (2013). Molecular Cell Biology. UK: Macmillan Higher Education.
2. Parrington, J. (2015). The Deeper Genome. Oxford: OUP.
3. Rodriguez‐Ezpeleta, N., Hakenberg, M., and Aransay, A.M. (eds.) (2012). Bioinformatics for High Throughput Sequencing. New York: Springer.
4. Wang, X. (2016). Next Generation Sequencing Data Analysis. Florida: CRC Press.
5. Watson, J. et al. (2014). Molecular Biology of the Gene Seventh Edition. New York: Cold Spring Harbour Press.

Websites

1. www.ukbiobank.ac.uk/2018/04/whole‐genome‐sequencing‐will‐transform‐the‐research‐landscape‐for‐a‐wide‐range‐of‐diseases
2. www.nature.com/news/human‐genome‐project‐twenty‐five‐years‐of‐big‐biology‐1.18436
3. en.wikipedia.org/wiki/List_of_sequence_alignment_software#Short‐Read_Sequence_Alignment
4. www.ensembl.org
5. www.ucsc.edu
6. www.targetvalidation.org
7. www.ensembl.org/Homo_sapiens/Tools/Blast
8. samtools.sourceforge.net
9. www.mlo‐online.com/long‐read‐sequencing
10. www.10xgenomics.com
11. www.sanger.ac.uk/htgt/wge
12. www.humancellatlas.org
13. www.gtexportal.org