Formula reference
Every quantitative result in the curriculum, with the chapter that derives it. Nothing here is asserted — each entry points at where the derivation lives, so you can check the assumptions rather than trusting the formula.
Assumptions are listed because in genetics the assumptions are usually the interesting part. A formula applied outside them gives a confident wrong answer.
Sequencing and data
| Quantity | Formula | Derived in |
|---|---|---|
| Phred quality | Q = −10 log₁₀(Pₑᵣᵣₒᵣ) | Ch 41 |
| Error from Phred | P = 10^(−Q/10) | Ch 41 |
| ASCII encoding | char = chr(Q + 33) — Phred+33; legacy data may be +64 | Ch 41 |
| depth | C = N · L / G — reads × length ÷ genome size | Ch 40 |
| Uncovered fraction | P(depth = 0) = e^(−C) — Poisson, Lander–Waterman | Ch 40 |
| Depth distribution | P(depth = k) = e^(−C) C^k / k! | Ch 40 |
| Sort contigs descending; N50 = length of the contig at which cumulative sum first reaches half the assembly total | Ch 43 | |
| NG50 | As N50, but the target is half the estimated genome size | Ch 43 |
Quick reference: Q20 = 1% error, Q30 = 0.1%, Q40 = 0.01%.
Real coverage is overdispersed relative to Poisson — repeats and GC bias mean the observed zero-coverage fraction exceeds e^(−C). Measured in lab-03: 0.85% observed against 0.24% predicted at 6.05×.
Mutation and repair
| Quantity | Value / formula | Derived in |
|---|---|---|
| Replication fidelity | ~10⁻¹⁰ per base per replication — the product of base selection (10⁻⁵) × (10⁻²) × (10⁻³) | Ch 04 |
| ~1.1–1.3 × 10⁻⁸ per base per generation | verified-facts | |
| per generation | μ × G ≈ 1.2 × 10⁻⁸ × 6.2 × 10⁹ ≈ 68–81 naive; ~60–70 observed (callable fraction) | Ch 16 |
| ≈ +1.3–1.5 de novo mutations per additional year of paternal age | Ch 16 |
The single most common unit error in the subject. Fidelity is per replication; the germline rate is per generation. They differ ~100× and are not comparable. A generation spans hundreds of divisions and also accumulates unrepaired chemical damage that was never a polymerase error.
Linkage and mapping
| Quantity | Formula | Derived in |
|---|---|---|
| RF = recombinants / total; 1% RF ≡ 1 cM | Ch 14 | |
| Ceiling | RF ≤ 0.5 always — multiple crossovers restore parental configurations | Ch 14 |
| Haldane mapping function | d = −½ ln(1 − 2·RF) Morgans — Poisson crossovers, no | Ch 14 |
| d = ¼ ln[(1 + 2·RF)/(1 − 2·RF)] — allows distance-dependent interference | Ch 14 | |
| Coefficient of coincidence | c.o.c. = observed DCO / expected DCO | Ch 14 |
| Interference | I = 1 − c.o.c. | Ch 14 |
| LOD = log₁₀(L() / L(no linkage)); threshold 3.0 | Ch 14 | |
| Cotransduction frequency | C(d) = (1 − d/L)³ for d < L — d the marker separation, L the phage headful; cube because the fragment must span both markers and each outer flank needs room for a crossover | Ch 20A §5 |
| Cotransduction distance | d = L(1 − C1/3) — the inverted form is the additive one; raw cotransduction frequencies must not be added | Ch 20A §5 |
| Time of entry (interrupted mating) | Position is an affine function of entry time; differences in entry time give map distance in minutes, absolute times do not — every curve is displaced by a few minutes of pair formation. E. coli map = 100 minutes | Ch 20A §3 |
Three-point cross: parentals are most frequent, double crossovers least. The gene that differs between a DCO class and its parental is the middle gene. Both DCO classes must be added into both interval distances — the commonest error in the calculation.
Both bacterial rulers saturate, and saturation is not a distance. Cotransduction returns C = 0 for every d ≥ L, so "not cotransducible" and "on the far side of the " are the same observation — the RF ≤ 0.5 ceiling above, in different clothing. At L ≈ 2 minutes, P1 transduction is blind beyond ~100 kb and bottoms out around 1–2 kb at the other end. Interrupted mating covers the whole chromosome but only to a minute; the two are used together, coarse then fine.
Human scale: ~3,400–3,500 cM autosomal genetic map over the ~2,875 Mb of autosome ⇒ ~1.2 cM/Mb average. Keep both halves of that ratio on the autosomes: dividing the autosomal map by the 3.1 Gb whole haploid genome is a common slip and understates the rate by ~7%. Recombination is in any case highly non-uniform (hotspots, centromeric suppression, ~1.6× longer female map).
Hardy–Weinberg and allele frequencies
| Quantity | Formula | Derived in |
|---|---|---|
| HWE frequencies | p² : 2pq : q², reached in one generation of random mating | Ch 26 |
| from genotypes | p = (2·N_AA + N_Aa) / 2N | Ch 26 |
| , rare | 2pq ≈ 2q; carriers outnumber affected by ≈ 2/q | Ch 26 |
| Male genotype frequency = allele frequency (); affected male:female ratio = 1/q | Ch 13 | |
| HWE chi-square | df = (genotypes) − () = 1 for biallelic — not k − 1 | Ch 12 |
Assumptions: random mating, no selection, no , no migration, infinite population.
HWE is a null model, not a prediction. In modern practice a departure is used as a genotyping-error filter, not as a biological discovery.
Selection, drift and effective size
| Quantity | Formula | Derived in |
|---|---|---|
| Selection against lethal | q′ = q/(1 + q); closed form qₙ = q₀/(1 + n q₀) | Ch 27 |
| Halving time | 1/q generations — each successive halving takes twice as long | Ch 27 |
| , recessive | q̂ = √(μ/s) | Ch 27 |
| Mutation–selection balance, | q̂ = μ/(hs) | Ch 27 |
| equilibrium | q̂ = s₁/(s₁ + s₂) — stable | Ch 27 |
| variance | Var(Δp) = pq/2N per generation | Ch 27 |
| Neutral probability | = current frequency p | Ch 27 |
| Mean time to fixation (neutral) | ≈ 4Nₑ generations | Ch 27 |
| Heterozygosity decay | Hₜ = H₀(1 − 1/2Nₑ)^t | Ch 27 |
| Nₑ, unequal sex ratio | Nₑ = 4N_f N_m/(N_f + N_m) | Ch 27 |
| Nₑ, fluctuating size | harmonic mean: Nₑ = n / Σ(1/Nᵢ) | Ch 27 |
| Nₑ, variance in family size | Nₑ ≈ (4N − 2)/(V_k + 2) — exceeds N when V_k < 2, which managed breeding exploits | Ch 35A §7 |
| Nₑ/N in wildlife | ≈ 0.1 — the three corrections above multiply; Frankham's comprehensive estimates average 0.10–0.11 across 102 species. A cross-species average over enormous variance, so it is a modelling assumption, not a constant | Ch 35A §7 |
| Inbreeding accumulation | ΔF = 1/(2Nₑ) per generation in a closed population — whatever the mating system, because the mate pool is finite | Ch 35A §7 |
| 50/500 rule | Nₑ ≥ 50 short-term (the size at which ΔF = 1%/generation); Nₑ ≥ 500 long-term (–drift balance for V_A). Frankham et al. 2014 argue 100/1000; unresolved — quote the derivation, not the threshold | Ch 35A §7 |
|Nₑs| is the most important inequality in population genetics. ≫1 ⇒ selection dominates; ≪1 ⇒ drift dominates and the variant behaves as if neutral. It is also why a small population cannot purge its own load: the small Nₑ that caused the inbreeding is exactly what blinds selection to the mildly deleterious that make up most of it — measured at under 1% across 119 captive pedigreed populations (Ch 35A §7).
Structure and inbreeding
| Quantity | Formula | Derived in |
|---|---|---|
| Inbreeding coefficient | F = probability two at a locus are identical by descent | Ch 28 |
| F from a loop | F = Σ (½)^(n₁+n₂+1) over all paths through common ancestors | Ch 28 |
| frequencies under inbreeding | p² + Fpq : 2pq(1 − F) : q² + Fpq | Ch 28 |
| F_ST | (H_T − H_S)/H_T | Ch 28 |
| F-statistic identity | (1 − F_IT) = (1 − F_IS)(1 − F_ST) | Ch 28 |
| Pooling differentiated subpopulations always produces a heterozygote deficit | Ch 28 |
Common F values: first cousins 1/16, half-sibs 1/8, uncle–niece 1/8, parent–offspring 1/4.
At a single locus, population structure and inbreeding are indistinguishable — both give a homozygote excess. Separating them requires many loci. Measured F_ST(AFR, EUR) = 0.065 on one 1 Mb region in lab-07.
Linkage disequilibrium
| Quantity | Formula | Derived in |
|---|---|---|
| D | D = p_AB − p_A·p_B | Ch 29 |
| D′ | D / D_max — normalised so the range does not depend on | Ch 29 |
| r² | D² / (p_A q_A p_B q_B) — the squared correlation between loci as 0/1 indicators | Ch 29 |
| Decay | Dₜ = D₀(1 − c)^t | Ch 29 |
| Power cost of tagging | Sample size scales as 1/r² when testing a tag rather than the causal variant | Ch 29 |
LD is not . Loci can be linked without LD, and transiently in LD without linkage.
Quantitative genetics
| Quantity | Formula | Derived in |
|---|---|---|
| Phenotypic variance | V_P = V_G + V_E + 2Cov(G,E) + V_GxE | Ch 30 |
| Genetic variance | V_G = V_A + V_D + V_I | Ch 30 |
| H² = V_G/V_P | Ch 31 | |
| h² = V_A/V_P — the one that predicts response to selection | Ch 31 | |
| Parent–offspring covariance | ½V_A | Ch 30 |
| Full-sib covariance | ½V_A + ¼V_D | Ch 30 |
| Falconer's formula | h² = 2(r_MZ − r_DZ) — exact only when V_D = V_I = 0; the full expression is h² = 2(r_MZ − r_DZ) − 1.5d² with d² = V_D/V_P, so the estimate is biased upward under dominance or | Ch 31 |
| R = h² · S | Ch 31 |
is a property of a population in an environment. It says nothing about an individual, does not mean "unchangeable", and licenses no inference about between-group differences. See the seeds-in-two-pots argument in Ch 31.
Molecular evolution
| Quantity | Formula | Derived in |
|---|---|---|
| Neutral substitution rate | k = μ — independent of population size (2Nμ new × 1/2N probability) | Ch 33 |
| distance | d = −¾ ln(1 − 4p/3) | Ch 33 |
| ω = (d_N per nonsynonymous site) / (d_S per synonymous site) | Ch 33 | |
| π = mean pairwise differences per site | Ch 33 | |
| Watterson's estimator | θ_W = S / Σ(1/i), i = 1…n−1 | Ch 33 |
| Tajima's D | ∝ (π − θ_W), scaled by its standard deviation | Ch 33 |
| ILS discordance | P( ≠ species tree) = ⅔·e^(−τ), τ = T/2N | Ch 34 |
| Patterson's D | D = (n_ABBA − n_BABA)/(n_ABBA + n_BABA) for (((P1,P2),P3),O); expectation 0 under ILS alone, which is symmetric. Significance by block jackknife — sites within a block are not independent | Ch 34, generalised in Ch 35A §5 |
| Combined | RI = 1 − Π(1 − bᵢ) — barriers act in sequence, so each sees only what the previous ones let through and an early barrier of a given strength always dominates the accounting | Ch 35A §2 |
| Dobzhansky–Muller snowball | Untested cross-lineage pairs = k(K − k) after K substitutions, = K²/4 when symmetric; E[incompatibilities] ≈ pK²/4 — quadratic in divergence, and with k = μ therefore quadratic in time | Ch 35A §3 |
| , dominance condition | cost ∝ 1 () against ∝ 2h (homogametic) ⟹ the heterogametic sex is worse off iff h < ½ | Ch 35A §4 |
| width | w = σ√(8/s) ⟹ s = 8(σ/w)² — Bazykin underdominance, width defined as inverse of the maximum gradient. Barton & Gale give ≈2.5σ/√s (endogenous), ≈1.7σ/√s (exogenous) | Ch 35A §5 |
| Neutral spread of a contact zone | width ~ σ√t — dispersal is diffusion, so it is the variance that adds per generation. Run this null before invoking selection | Ch 35A §5 |
Interpretation: ω < 1 purifying, = 1 neutral, > 1 positive. Gene-wide ω rarely exceeds 1 even under real positive selection — use branch-site or site models.
Tajima's D confounds selection with demography. Negative D is consistent with a or population expansion. Corrected distances can exceed 1.0 substitution per site even though observed p-distance saturates at 0.75 — see lab-10, where human–mouse mtDNA distance is 1.218.
The √s scaling of a cline is robust; the constant is a model choice you must declare. Inverting w ~ σ/√s as if the prefactor were 1 understates selection eightfold — the error worked through on the Bombina transect in Ch 35A, where w = 6.05 km and σ = 0.99 km/generation give s ≈ 0.21, not 0.027. And s goes as the square of σ, so a 20% error in dispersal is a 44% error in selection.
Statistical genomics
| Quantity | Formula | Derived in |
|---|---|---|
| significance threshold | 5 × 10⁻⁸ ≈ 0.05 / 10⁶ effectively independent tests | Ch 51 |
| λ = median(observed χ²) / median(null χ²) = median χ²/0.4549 for 1 df | Ch 51 | |
| Additive coding | dosage ∈ {0, 1, 2}, plus covariates (PCs) | Ch 51 |
| PRS = Σ βᵢ · dosageᵢ, weights shrunk toward zero | Ch 53 | |
| Benjamini–Hochberg | Reject H(ᵢ) where p(ᵢ) ≤ (i/m)·α | Ch 47 |
| Bayesian risk | posterior ∝ prior × conditional; tabulate prior / conditional / joint / posterior | Ch 12, Ch 15 |
λ alone cannot distinguish confounding from true polygenicity. Use the intercept: raised intercept ⇒ confounding; raised slope with intercept ≈ 1 ⇒ real . Demonstrated in lab-08, where a with no genetic cause produced 702 genome-wide significant hits at λ = 18.07, falling to 0 hits at λ = 1.14 once ancestry PCs were included.
Constants
| Quantity | Value | Source |
|---|---|---|
| Human haploid genome | 3.1 Gb (: 3,117,292,070 bp) | verified-facts |
| Protein-coding genes | 19,442 (GENCODE 50) | verified-facts |
| Non-coding genes | 58,195 — never compute as total − coding | verified-facts |
| content | ~46% | verified-facts |
| B-DNA rise per bp | 0.34 nm | Ch 02 |
| B-DNA bp per turn | ~10.5 | Ch 02 |
| 147 bp; ~166 bp chromatosome with H1 | Ch 03 | |
| Human genetic map | ~3,400–3,500 cM sex-averaged, autosomal (male ~2,600–2,700; female ~4,200–4,400) | Ch 14 |
| Human autosome length | ~2,875 Mb — the denominator for cM/Mb, not the 3.1 Gb whole genome | Ch 13 |
| Human generation time | ~27 years (sex-averaged) | verified-facts |
| Human Nₑ (historical) | ~10,000 | Ch 27 |
| / amino acids | 64 / 20 + stop | Ch 07 |