Glossary

Every defined term in the course, with links to where it is taught

Type in the box to filter. Each entry links back to the lecture section(s) where the term is defined or used. Quarto’s search (top right) also searches the full text of every lecture.

A–C

Additive model

The standard GWAS model: phenotype regressed on allele dosage (0/1/2), so each extra copy adds the same effect β. Additive/dominant/recessive describe the genotype–phenotype relationship, not the allele. L4 §4.4

Allele

An alternative version (DNA sequence) at a particular locus. L3 §3.6 · L4 §4.2

Allele frequency / MAF

How common an allele is in a population. MAF (minor allele frequency) is the frequency of the less common allele. L4 §4.2

Alternative splicing

One gene’s transcript spliced into several different mRNA/protein variants; one reason proteins are not 1-to-1 with genes. L8 §8.1

AML (acute myeloid leukemia)

A non-solid blood cancer driven by somatic mutations in bone marrow stem cells; the course’s example of somatic mutation diagnostics with NGS and VAF. L3 §3.4

Antibody-based assay

ELISA, clinical immunoassays, Western blot: an antibody binds its target protein; targeted and quantitative with calibration standards. L8 §8.4

Assay binding artifact

A coding variant changes the protein’s structure so the antibody/aptamer binds worse, giving a falsely low measured level. A protein-MR pitfall. L8 §8.11

Assortative mating

Non-random mate choice related to a trait; a threat to the MR independence assumption. L6 §6.5

AUC (area under the ROC curve)

Probability that a random case gets a higher predicted risk than a random control; 0.5 = chance, 1 = perfect. Used to ask whether a PRS adds discrimination. L9 §9.13

Background correction (normexp)

Within-array correction modelling observed intensity as exponential signal + normal background noise; gives positive values and better low-intensity estimates. L7 §7.5

Batch effect

A shared technical artifact among samples processed together (plate, reagent lot, run, day, instrument, operator). Prevent by randomizing/balancing; correct by normalization or batch covariates. L7 §7.4 · L8 §8.6

Benjamini–Hochberg (BH)

FDR-controlling procedure: sort p-values, find the largest k with p₍ₖ₎ ≤ (k/m)Q, call ranks 1…k significant. Adjusted p = running minimum of p₍ᵢ₎·m/i from the bottom. L7 §7.13 · L7 §7.13b

Biomarker

Any measurable biological characteristic informative about a process, exposure, disease or treatment response. Five roles: exposure, process, disease/injury, future risk, treatment response. Not necessarily causal. L1 §1.3b

Bonferroni correction

FWER control: reject if p ≤ α/m (p_adj = p × m). Valid under any dependence but very strict for large m. L7 §7.12 · L4 §4.7

Bradford Hill criteria

Strength, Consistency, Specificity, Temporality, Biological gradient, Plausibility, Coherence, Experiment, Analogy. L6 §6.2

Calibration

Agreement between predicted and observed risks (predicted 10% → about 10% get the disease). L9 §9.13

Central dogma

The flow of genetic information DNA → RNA → protein. L1 §1.4 · L2 §2.2

Chromosomal alterations

Large-scale changes: deletion, duplication, inversion, insertion, translocation. L2 §2.4b

Circular reasoning (double dipping)

Using the same data to define groups (e.g. clusters) and then to test differences between those groups; inflates significance. L7 §7.16b

cis-pQTL

A pQTL located near the gene encoding the protein; direct interpretation, preferred MR instrument. L8 §8.9 · L4 §4.10

Clumping + thresholding (C+T)

Classical PRS method (“pruning and thresholding”): keep one lead SNP per locus, include SNPs with P below a threshold. L9 §9.12

Codon

Three nucleotides specifying one amino acid (or stop). 64 codons for 20 amino acids. L1 §1.4b · L2 §2.3

Coding strand

The non-template DNA strand; same sequence as the mRNA with T instead of U. L2 §2.2

Colocalization

Checks whether a pQTL and a disease GWAS signal at one locus share the same causal variant by comparing their association patterns across the region. L8 §8.11

Complementary base pairing

A pairs with T (U in RNA), C pairs with G; the basis of replication and transcription. L1 §1.4b

Conditional analysis

Adding the lead SNP as a covariate and retesting other SNPs to find independent signals. L5 §5.4

Confounding

A third factor affecting both exposure and outcome (e.g. smoking → LDL-C and CHD). L6 §6.4 · L8 §8.8

Covariates (in GWAS)

Age, sex, BMI, genetic PCs, technical variables (batch, center). β is conditional on them. L4 §4.5

Credible set

A set of candidate causal SNPs chosen from association strength and LD structure. L5 §5.4

D–H

dbSNP / rs number

NCBI database of short variants; each gets a stable rs identifier (e.g. rs137852912 = PCSK9 D374Y). L3 §3.3c

Discovery (base) and target sample

In PRS analysis: the discovery GWAS supplies the β estimates; the independent target sample supplies genotypes, phenotype and covariates. L9 §9.11

Dominant / recessive allele

Dominant affects the phenotype with at least one copy; recessive only with two. L3 §3.6

Empirical Bayes / variance shrinkage

Sharing information across features: each variance is pulled toward a common value. Basis of limma’s moderated t. L7 §7.7 · L7 §7.8

Exclusion restriction

MR assumption 3: the variant affects the outcome only through the exposure. Broken by horizontal pleiotropy. L6 §6.5

F-statistic (instrument strength)

F = (β_GX / SE)² for one SNP; F < 10 marks a weak instrument. L9 §9.3

FDR (false discovery rate)

Expected proportion of false positives among the tests called significant. Controlled by BH. L7 §7.11

Frameshift

An indel whose length is not a multiple of 3, shifting all downstream codons. L2 §2.4b

FWER (family-wise error rate)

Probability of at least one false positive among all tests. Controlled by Bonferroni. L7 §7.11

Gain / loss of function (GoF / LoF)

A mutation making the protein work better (D374Y in PCSK9) or worse (p.Tyr142*, p.Arg97del). L3 §3.3

GalNAc / ASGPR

Sugar linked to the siRNA in Leqvio that binds ASGPR receptors on hepatocytes: an “address label” to the liver. L3 §3.2b

Gene expression

Transcription + translation; regulated at chromatin, transcription, RNA processing, mRNA degradation and post-translational levels. L1 §1.4 · L2 §2.3

Genome build (GRCh37/hg19, GRCh38)

Versions of the human reference genome; the same variant has different coordinates in different builds. HG18/HG19 are older names. L3 §3.3c · L4 §4.6

Genome-wide significance

p < 5 × 10⁻⁸ ≈ 0.05 / 1,000,000. L4 §4.7

Genotype / phenotype

Genotype = the alleles an individual carries; phenotype = the observable trait. L3 §3.6

Genotyping

Determining alleles at already-known variant positions (TaqMan, SNP arrays), as opposed to sequencing. L3 §3.5 · L3 §3.8

Germline mutation

Arises in germ cells, present in every cell, heritable. L3 §3.4

GWAS

Genome-wide association study: one regression per SNP for ~1 million SNPs. Gives associations, not causality. L3 §3.8 · L4 §4.6 · L5 §5.2

Haplotype

A combination of alleles at two or more loci on the same chromosome copy. L3 §3.7

Hardy–Weinberg equilibrium (HWE)

With allele frequencies p and q: P(AA) = p², P(Aa) = 2pq, P(aa) = q². Deviations are a genotyping QC check. L3 §3.9 · L4 §4.2

Harmonisation

Aligning alleles so the exposure and outcome β refer to the same effect allele (flip the sign when swapped; handle strand and palindromic SNPs). L9 §9.5

HGVS nomenclature

Standard variant notation: reference (NM_ mRNA, NC_ genomic, NP_ protein) + coordinate type (c., g., p.) + change (e.g. c.1120G>T, p.Asp374Tyr). L3 §3.3b

Homozygous / heterozygous

Two identical alleles (BB, bb) vs two different alleles (Bb). L3 §3.6

Horizontal pleiotropy

A variant affects the outcome through a pathway independent of the exposure; violates the exclusion restriction. L6 §6.6 · L8 §8.11

I–M

Imputation

Predicting genotypes at unmeasured SNPs from measured SNPs using LD and a reference panel. L6 §6.1

In-frame indel

An insertion/deletion of a multiple of 3 nucleotides; frame preserved (e.g. PCSK9 p.Arg97del). L2 §2.4b

Independence (MR assumption)

MR assumption 2: the variant is not associated with confounders. Threats: population stratification, assortative mating, selection. L6 §6.5

Instrumental variable

A variable affecting the exposure but related to the outcome only through it; genetic variants are natural instruments. L6 §6.3

IVW (inverse-variance weighted) MR

Weighted regression of β-outcome on β-exposure through the origin = weighted average of Wald ratios. L6 §6.7 · L9 §9.6

LD clumping

Reducing each locus to one approximately independent lead SNP (e.g. r² < 0.01); done before MR. L5 §5.4b · L9 §9.3

LDL / LDLR

Low-density lipoprotein carries cholesterol; the LDL receptor on hepatocytes clears it. L1 §1.2

Lead SNP

The SNP with the smallest p-value in a locus; not necessarily causal. L4 §4.8 · L5 §5.3

limma / moderated t-statistic

R package using empirical Bayes variance shrinkage to give stable t-statistics with few samples. L7 §7.7

Linkage

Loci close on the same chromosome are inherited together. L3 §3.7

Linkage disequilibrium (LD)

Non-random association of alleles at different loci in a population; observed ≠ expected haplotype frequency. Measured by r² (or D). L3 §3.7 · L4 §4.8

Locus

A position on a chromosome; in GWAS, a region containing associated variants (one Manhattan peak). L3 §3.6 · L4 §4.10

Log₂ fold change

log₂(group 1 / group 2): 0 = no change, +1 = 2× higher, −1 = 2× lower. L7 §7.9

Manhattan plot

−log₁₀(p) against genomic position; peaks show where associations are, not which variant is causal. L4 §4.7

Mass spectrometry

Trypsin digestion → liquid chromatography → electrospray ionization → mass measurement; detects isoforms, allows discovery. L8 §8.4

Meiosis / mitosis

Meiosis: 4 different gametes with 23 chromosomes. Mitosis: 2 identical cells with 46. L1 §1.4c

Mendel’s laws

Dominance, Segregation, Independent Assortment (broken by linkage). L3 §3.6

Mendelian randomization (MR)

Using genetic variants as instruments to estimate causal effects of an exposure on an outcome. L6 §6.3 · L8 §8.10

Metabolite

A small molecule produced, modified or consumed in biochemical reactions (e.g. glucose). L1 §1.3

miRNA (microRNA)

~22 nt genome-encoded non-coding RNA; binds targets imperfectly and regulates many mRNAs. L3 §3.2 · L7 §7.2

Missense / nonsense / silent

Substitution giving a different amino acid / a premature stop / the same amino acid. L2 §2.4

Monoclonal antibody

An engineered protein binding one specific target; Repatha and Praluent bind PCSK9 protein. L3 §3.2b

MR-Egger

MR regression with a free intercept; a non-zero intercept signals directional horizontal pleiotropy. Less precise than IVW. L6 §6.8

mRNA / rRNA / tRNA

Messenger RNA carries the code; ribosomal RNA builds ribosomes; transfer RNA brings amino acids. L2 §2.1

Multivariable MR

MR with several related exposures in one model (e.g. apoB, LDL-C and remnant cholesterol), so each causal estimate is adjusted for the others. L9 §9.14

Mutation

A change in the DNA sequence (L2’s definition). Point mutations (silent, missense, nonsense), indels (frameshift, in-frame) and chromosomal alterations; germline or somatic; gain or loss of function. In practice largely interchangeable with variant, which is the more neutral term. L2 §2.4 · L2 §2.4b · L3 §3.3 · L3 §3.4

Multiple testing

Many tests at α = 0.05 give many false positives (955 × 0.05 ≈ 48; 1,000,000 × 0.05 = 50,000). L7 §7.10 · L4 §4.7

N–R

NGS (next-generation sequencing)

Massively parallel sequencing: fast and cheap, short reads (~50–300 nt). L3 §3.5

NPX (Normalized Protein eXpression)

Olink/SomaScan relative abundance on a log₂ scale; 1 NPX = a doubling. L8 §8.5

Olink / SomaScan

High-throughput affinity proteomics platforms for large cohorts. L8 §8.4

OncomiR

A miRNA that promotes cancer (e.g. miR-1246, miR-663a). L7 §7.2 · L7 §7.16

Odds ratio (per allele)

OR = exp(β) from logistic regression; β = −0.16 → OR 0.85 (one copy), 0.73 (two copies). L4 §4.3

Palindromic SNP

A/T or C/G SNP whose complement is the same allele pair, so strand cannot be read from the letters; resolved by allele frequency or excluded. L9 §9.5

PCR

Polymerase chain reaction: in vitro amplification of a DNA region using heat-stable Taq polymerase and primers. L2 §2.6

PCSK9

Protein promoting LDLR degradation; the course’s running drug-target example. L1 §1.2 · L3 §3.3 · L8 §8.12

Polygenic risk score (PRS)

Sum over SNPs of allele dosage × GWAS effect size for one individual; a relative measure of genetic liability, not a probability. L9 §9.10

Population stratification

Allele frequency and phenotype both differ between ancestry groups, creating spurious associations. L4 §4.9

Pre-analytical variation

Variation before the assay: fasting, time of day, tube type, time to centrifugation, storage, freeze–thaw, hemolysis. L8 §8.6

Primer

Short DNA sequence where synthesis starts; DNA polymerase needs one, RNA polymerase does not. L2 §2.2 · L2 §2.6

Protein structure (4 levels)

Primary (sequence), secondary (helices, sheets), tertiary (3D fold), quaternary (several chains). L1 §1.4

Proteome

All proteins in a sample; more dynamic and complex than the genome. L8 §8.1

Proxy SNP

A SNP in very strong LD (r² ≈ 1) that stands in for a missing SNP. L6 §6.1

Post-translational modification (PTM)

Chemical changes after translation, e.g. phosphorylation, glycosylation, cleavage. L8 §8.1

pQTL (protein quantitative trait locus)

A SNP associated with a protein’s abundance; cis (near the gene) or trans (> 500 kb away). L8 §8.9

PWAS

Proteome-wide association study: one regression per protein. L8 §8.7

QQ plot

Observed vs expected p-values; deviation can mean confounding/batch or high polygenicity. L4 §4.7

QTL

Quantitative trait locus: a region where genetic variation is associated with a quantitative trait (expression, protein, metabolite). L4 §4.10

Quantile normalization

Between-array normalization making all distributions identical while keeping within-array order. L7 §7.5

r² (LD)

Squared genotype correlation between two SNPs: > 0.8 very strong, > 0.5 moderate–strong, < 0.1 weak, ≈ 0 independent. L4 §4.8 · L5 §5.4b

Recombination / recombination frequency

Exchange of segments between homologous chromosomes in meiosis; r = recombinants / total, 0–50%. L3 §3.7

Relevance (MR assumption)

MR assumption 1: the variant is strongly associated with the exposure; threat: weak instruments. L6 §6.5

Remnant cholesterol (TRL/remnant-C)

Cholesterol in triglyceride-rich lipoproteins and their partly lipolysed remnants; computed as non-HDL-C − LDL-C. L9 §9.14

S–Z

Sample overlap

Individuals present in both GWAS (MR) or in both discovery and target data (PRS); biases MR toward the observational estimate and inflates PRS accuracy. L9 §9.8 · L9 §9.11

Sanger sequencing

The original sequencing method: accurate, lower throughput. L3 §3.5

siRNA

~21–23 nt RNA from double-stranded RNA; near-perfect match cleaves one target mRNA (Leqvio). L3 §3.2

SNP / SNV

Single nucleotide variant; called a SNP (polymorphism) when common in the population. L2 §2.4 · L4 §4.2

SNP array

Chip with allele-specific probes genotyping up to millions of known SNPs by hybridization and fluorescence. L3 §3.8

snRNA

Small nuclear RNA involved in splicing. L3 §3.1

Somatic mutation

Acquired during life, confined to a tissue, not heritable. L3 §3.4

Summary statistics

Per-SNP GWAS results (SNP, CHR, BP, A1, A2, EAF, β, SE, P, N) with no individual-level data. L4 §4.6 · L9 §9.2

Supervised / unsupervised analysis

Using predefined group labels (differential expression) vs finding groups from the data (clustering). L7 §7.6

Taq polymerase

Heat-stable DNA polymerase used in PCR. L2 §2.6

Template strand

The DNA strand read by RNA polymerase (non-coding strand). L2 §2.2

Transcription / translation

DNA → mRNA (in the nucleus) / mRNA → protein at the ribosome (initiation, elongation, termination at UAA, UAG, UGA). L2 §2.2 · L2 §2.3

Transcription factor

Protein binding DNA to raise or lower transcription. L2 §2.2

trans-pQTL

pQTL far (> 500 kb or other chromosome) from the encoding gene; indirect, more pleiotropy risk. L8 §8.9

Troponin

Released by damaged heart muscle; a marker of injury, predictive but not causal. L1 §1.3b · L8 §8.8

Two-sample MR

MR with SNP → exposure and SNP → outcome effects taken from two different GWAS; needs only summary statistics. L9 §9.1

VAF (variant allele frequency)

Fraction of reads/cells in one patient’s sample carrying a mutation. L3 §3.4

Variant (genetic variant)

Any place where a person’s DNA sequence differs from the reference genome (or from other people). The neutral umbrella term: it covers SNVs/SNPs, small insertions and deletions (indels) and larger structural changes, whether common or rare, harmful or harmless. It is the word used in HGVS notation, dbSNP (rs numbers), GWAS and MR (“genetic variant as instrument”). Mutation usually stresses the change event or a rare, often disease-related change; polymorphism (the P in SNP) a variant that is common in the population. L2 §2.4 · L3 §3.3b · L3 §3.3c · L4 §4.2 · L6 §6.3

Vertical pleiotropy

The variant affects the outcome through the exposure itself; not a problem for MR. L6 §6.6

Wald ratio

MR estimate with one instrument: β_outcome / β_exposure (β_GO / β_GP in protein MR). L6 §6.7 · L8 §8.10 · L9 §9.6

Weak instrument

A SNP explaining little of the exposure (F < 10); biases MR toward the null (no overlap) or toward the confounded estimate (overlap). L9 §9.8

Weighted median

MR estimator valid if at least ~50% of instrument weight is valid. L6 §6.8

Winner’s curse

Effect sizes of newly discovered variants tend to be overestimated in the GWAS that found them; in MR this biases toward the null. L9 §9.8

z-score

(x − mean) / SD; puts proteins on a common scale, β per SD. L8 §8.6 · L8 §8.7