1 Molecular biology for bioinformaticians

Every genetic dataset you will touch is a measurement of one layer of a single pipeline: DNA is copied into RNA, and RNA is translated into protein. A genotype file records the DNA layer, an RNA-seq matrix records the RNA layer, and a methylation array records a control layer on top. Know which layer your data measures and most downstream choices follow.

1.1 The central dogma: DNA makes RNA makes protein

The flow of information in a cell runs in one main direction. DNA stores it, RNA carries it, and proteins do most of the work.

flowchart LR
    DNA["DNA<br/>genome"] -- "transcription" --> PRE["pre-mRNA"]
    PRE -- "splicing" --> MRNA["mature mRNA"]
    MRNA -- "translation" --> PROT["Protein"]
    DNA -. "replication" .-> DNA
    MIR["microRNA"] -. "represses" .-> MRNA
    EPI["Epigenetic marks"] -. "control access" .-> DNA

DNA is a double helix of two antiparallel strands. Each strand is a chain of four bases: adenine (A), cytosine (C), guanine (G), and thymine (T). A pairs with T, and C pairs with G.
Transcription copies a stretch of DNA, a gene, into RNA. RNA uses uracil (U) instead of T.
Splicing removes introns from the precursor RNA and joins the exons into a mature messenger RNA (mRNA).
Translation reads the mRNA three bases at a time. Each three-base codon specifies one amino acid or a stop signal.

The human genome holds about 3.1 billion base pairs per haploid copy, on 22 autosomes plus the X and Y chromosomes, and the small mitochondrial genome. Only about 1–2% of it codes for protein. Roughly 20,000 protein-coding genes sit among many more non-coding genes and regulatory elements.

1.2 Genome coordinates: chromosome, position, build

A location in the genome is a chromosome plus a position, written like chr1:234 or 1:234. That pair is only meaningful together with the genome build (reference assembly) it refers to.

Build	Also called	Released	Where you meet it
GRCh37	hg19	2009	Older arrays, many published summary statistics
GRCh38	hg38	2013	Current sequencing, gnomAD v3+, most new biobank releases
T2T-CHM13	—	2022	First gap-free assembly; growing use in research

The same variant has different positions in different builds. Converting coordinates between builds is called liftover (UCSC liftOver, CrossMap, Picard LiftoverVcf).

File formats also differ in how they count. You will see the formats themselves in Chapter 6.

Convention	Used by	First base is	A one-base variant at base 100
1-based, closed	VCF, GFF/GTF, SAM, PLINK `.bim`	1	`100`
0-based, half-open	BED, BAM internally, UCSC tables	0	`99 100`

Pitfall

Mixing GRCh37 and GRCh38 coordinates, or a 0-based and a 1-based file, produces no error. It produces joins that silently miss or mismatch variants. Record the build in every file name or header, and check it before any merge.

Strand matters too. Positions refer to the forward (+) strand of the reference. A gene can lie on either strand, and older array manifests sometimes report alleles on the reverse strand. A/T and C/G SNPs are ambiguous, because their complement is the same pair of letters. Strand problems are handled in Chapter 9.

1.3 Bases, alleles, and genotypes

At a variable position, each version of the sequence is an allele. The allele in the reference genome is the reference allele (REF); the other is the alternate allele (ALT). REF is not “normal” or “ancestral”; it is simply what the reference assembly happened to contain.

Humans are diploid: you carry two copies of each autosome, one from each parent. Your genotype at a site is the pair of alleles you carry.

Genotype	Description	VCF `GT`	Alternate allele count
A/A	Homozygous reference	`0/0`	0
A/G	Heterozygous	`0/1`	1
G/G	Homozygous alternate	`1/1`	2
unknown	Missing call	`./.`	NA

In VCF, 0 means REF and 1 means the first ALT. A slash (0/1) means the genotype is unphased: you know the two alleles but not which parental chromosome carries each. A pipe (0|1) means phased. Chapter 2 explains why that matters.

1.4 Dosage: genotypes as numbers

Association models need numbers, not letters. The standard encoding counts copies of one chosen allele: 0, 1, or 2. This is the additive coding, and the counted allele is the effect allele whose beta or odds ratio you will later report.

When genotypes are imputed rather than directly measured, the imputation software (Minimac4, IMPUTE5, Beagle) gives a probability for each of the three genotypes. The dosage is the expected allele count:

\[\text{DS} = 0 \cdot P(AA) + 1 \cdot P(AG) + 2 \cdot P(GG)\]

It is a continuous number between 0 and 2.

Sample	P(AA)	P(AG)	P(GG)	Dosage (G)	Reading
S1	0.05	0.25	0.70	1.65	Probably GG
S2	0.50	0.45	0.05	0.55	AA or AG; uncertain
S3	0.00	1.00	0.00	1.00	Confidently AG
S4	0.80	0.20	0.00	0.20	Probably AA

In a VCF, the same information sits in the DS and GP FORMAT fields:

#CHROM  POS     ID     REF  ALT  QUAL  FILTER  INFO  FORMAT    S1
1       123456  rs123  A    G    .     PASS    .     GT:DS:GP  1/1:1.65:0.05,0.25,0.70

Using dosages instead of rounding to hard calls keeps the imputation uncertainty in the model and usually gains power. PLINK 2, SAIGE, REGENIE, BOLT-LMM, and Hail all accept dosages. The PLINK 1 binary format (.bed) stores only hard calls, so converting dosages to it throws information away. Chapter 6 covers BGEN and PGEN, which store dosages.

Pitfall

“Dosage of what?” is the first question to ask. Some tools count ALT, some count the minor allele, and PLINK 1.9 counts its A1 allele, which defaults to the minor one. The same SNP can be coded 0/1/2 in one file and 2/1/0 in another, flipping the sign of every effect.

1.5 Genes, transcripts, and the transcriptome

A gene is not one sequence but a template for several. A typical protein-coding gene has a promoter, a 5′ untranslated region (UTR), exons separated by introns, and a 3′ UTR. Alternative splicing combines exons in different ways, so one gene can yield several transcripts (isoforms) and several proteins.

The transcriptome is the full set of RNA molecules in a cell or tissue at one moment. It includes:

mRNA, which is translated into protein;
ribosomal and transfer RNA (rRNA, tRNA), the translation machinery;
long non-coding RNA (lncRNA, over 200 nucleotides), with mostly regulatory roles;
small RNAs, such as microRNA (miRNA), small interfering RNA (siRNA), and piwi-interacting RNA (piRNA).

Unlike the genome, the transcriptome is dynamic. It changes with cell type, developmental stage, time of day, disease, and drug exposure. That is why RNA-seq is used to find differentially expressed genes, define cell types, and link variants to expression (eQTL studies, used in Chapter 15).

1.6 Same DNA, different cells

Almost every cell in your body carries the same genome, yet a neuron and a liver cell look and behave nothing alike. The difference lies in which genes each cell expresses, and how much. Several layers of control stack on top of each other.

Layer	Mechanism	Effect on cell identity
Transcriptional	Transcription factors bind promoters and enhancers	Each cell type runs its own set of master regulators
Epigenetic	DNA methylation, histone marks, chromatin accessibility	Locks expression states in and passes them to daughter cells
RNA processing	Alternative splicing, alternative polyadenylation	Same gene, different proteins in different tissues
Post-transcriptional	microRNAs and RNA-binding proteins	Fine-tunes how much protein each mRNA yields
Signaling	Hormones, cell contact, environment	Tells a cell where it is and what to become

This is also why tissue matters in genetics. A variant can change expression in brain but not in blood, so expression and regulatory annotations must come from a relevant tissue.

1.7 Epigenetics

Epigenetics is the set of heritable changes in gene activity that do not change the DNA sequence. “Heritable” here mainly means passed from a cell to its daughter cells. Inheritance across human generations is debated and, at most, limited.

DNA methylation: a methyl group is added to cytosine, mostly at CpG sites (a C followed by a G). Dense methylation at a promoter usually silences the gene. Arrays such as the Illumina EPIC measure hundreds of thousands of CpGs; bisulfite sequencing measures them genome-wide.
Histone modifications: DNA wraps around histone proteins. Chemical marks on histone tails (acetylation, methylation, phosphorylation) loosen or tighten that packing. H3K27ac, for example, marks active enhancers; H3K27me3 marks repressed regions.
Chromatin accessibility: open chromatin can be read; closed chromatin cannot. ATAC-seq maps open regions.
Non-coding RNA: lncRNAs and miRNAs help direct where the other mechanisms act.

Epigenetic marks respond to age, diet, smoking, and other exposures. They also differ by cell type, which is the main confounder in methylation studies of mixed tissues such as whole blood.

1.8 MicroRNA

MicroRNAs are short non-coding RNAs, about 22 nucleotides long, that reduce how much protein a target mRNA produces. They act as dimmers rather than on/off switches.

Their life cycle is well defined. A long primary transcript (pri-miRNA) is cut in the nucleus into a hairpin (pre-miRNA). The enzyme Dicer trims it in the cytoplasm, and one strand is loaded into an Argonaute protein. The resulting complex binds target mRNAs, mostly in their 3′ UTR, and causes their decay or blocks translation.

Binding needs only a short match: the seed region, nucleotides 2–8 of the miRNA. One miRNA can therefore regulate hundreds of genes, and one gene can be targeted by many miRNAs. miRNAs also cross-talk with epigenetics: they target DNA methyltransferases and histone modifiers, and methylation of their own promoters can silence them.

miRNAs are stable in blood and other body fluids, which makes them attractive biomarkers of disease and environmental exposure. Profiling options include qRT-PCR, microarrays, and small-RNA sequencing, each with trade-offs in sensitivity and discovery (Pritchard, Cheng and Tewari, 2012, Nature Reviews Genetics). Reference names and sequences live in miRBase.

In practice

In miRNA studies, normalization and sample handling drive results more than the statistics do. Hemolysis releases red-cell miRNAs into serum and plasma, and platelet contamination does the same. Check for these before trusting a “biomarker”.

1.9 Extracellular vesicles

Cells release small membrane-bound particles, extracellular vesicles (EVs), that carry proteins, lipids, and RNA, including miRNAs, from one cell to another. Their membrane protects that cargo from degradation in blood.

Type	Typical size	Origin
Exosomes (small EVs)	about 30–150 nm	Inner vesicles of endosomes, released when the endosome fuses with the cell membrane
Microvesicles	about 100–1,000 nm	Bud directly off the plasma membrane
Apoptotic bodies	about 1–5 µm	Fragments of dying cells

Sizes overlap, and most isolation methods cannot separate the types cleanly. Current guidelines therefore recommend the generic term “extracellular vesicle” unless the origin is shown (Théry et al., 2018, MISEV2018, Journal of Extracellular Vesicles).

Because EVs carry markers of the cell they came from, they offer a possible “liquid biopsy” of tissues you cannot sample directly. Neuronally enriched EVs are the leading example. Researchers capture vesicles from serum or plasma using antibodies against neuronal surface proteins, most often L1CAM, and measure proteins such as tau, amyloid-β, or α-synuclein. The aim is a blood test that tracks Alzheimer’s disease, Parkinson’s disease, or brain injury.

The idea is promising but not settled. L1CAM is not exclusive to neurons, and one study found that it is mostly present in soluble form in plasma rather than on vesicles (Norman et al., 2021, Nature Methods). Treat neuronal-EV results as research findings, not validated diagnostics.

1.10 What this means for your data

Each molecular layer becomes a different kind of dataset, with its own formats and statistics.

Layer	Typical assay	Typical data	Where in this book
DNA sequence	Genotyping array, exome, genome	VCF, PLINK, BGEN; genotypes 0/1/2	Chapter 5, Chapter 6, Chapter 11
Epigenome	Methylation array, ATAC-seq	Beta values, peaks	Chapter 15 (annotation)
Transcriptome	RNA-seq, small-RNA-seq	Count matrices	Chapter 3 (count distributions)
Protein and EV cargo	Immunoassays, proteomics	Concentrations	Chapter 4

Try it

Look up the variant rs4244285 in Ensembl and in the GRCh37 archive (grch37.ensembl.org). Write down its chromosome, its position in both builds, its REF and ALT alleles, and the gene it falls in. Then explain, in one sentence, why a file of positions without a stated build is incomplete. You will meet this variant again in Chapter 2 and Chapter 16.