[Verse 1] Flip the genome like a coin, binomial counts the reads, Poisson catches rare mutations scattered in the weeds, Negative binomial when the variance runs too wide, Beta's the conjugate that keeps the prior on your side, Likelihood's the story that the data wants to tell, Maximize it, climb the hill, till the slope levels well, Bayes flips the conditional, prior meets the odds, Posterior's the answer, not from magic — just from logs. [Chorus] Prior times likelihood, divided by the evidence you find, That's posterior — Bayes rewiring what you thought you'd mined, One-step memory, the Markov chain forgets the rest, Forward adds the maybes, Viterbi picks the best. [Verse 2] Markov chain's got amnesia, only last state matters most, Hidden Markov hides the truth behind an emission ghost, Forward algorithm sums the paths that could've made the sound, Viterbi carves the single trail, the max path homeward bound, Baum-Welch trains it blindfold, no labels, just the churn, Guess it, weigh it, sharpen it — repeat until it learns, Exon, intron, splice site — gene finders walk the seam, CNV segments gain from loss, profile HMMs thread the theme. [Chorus] Prior times likelihood, divided by the evidence you find, That's posterior — Bayes rewiring what you thought you'd mined, One-step memory, the Markov chain forgets the rest, Forward adds the maybes, Viterbi picks the best. [Verse 3] Entropy counts the bits of surprise inside a site, Low where columns are conserved, high where anything's alright, KL divergence clocks how far two distributions drift, Mutual information links a motif to its cognate shift, Null versus story you hoped true — that's the hypothesis game, Bonferroni cuts the herd, guards against the false-claim shame, Benjamini-Hochberg ranks and lets discovery breathe, GLMs bend the fit to Poisson, binomial, past belief, Mixed models add the random slope for batch and lab and day, Permute the labels, bootstrap the sample, let the noise give way, EM alternates forever — expect it, then maximize, Hidden variables surface like a truth behind disguise. [Verse 4] PCA rotates the cloud to find the widest thread, SVD splits the matrix into rotate, scale, and spread, NMF stays strictly positive — signatures untangle clean, Random forests vote in crowds, each tree a different scene, CNNs slide a kernel, learning motifs pixel-wise, Transformers weigh attention, every base sees every prize, Lasso zeroes out the weights, ridge just shrinks them thin, Split by chromosome, not sequence — or your leakage sneaks back in.
← Module 14 — Reproducibility, Scale & Capstone (Weeks 27–30) | Appendix B — Molecular Biology Fast Track (for CS/math backgrounds) →