[Verse 1] (Male) Pileup stacked like witnesses, quality scores testifying, each read's a juror, Phred's the judge decrying, genotype likelihood — the data given the type, multiply the errors till the posterior's ripe, Bayes walks in with a prior on his shoulder, Hardy-Weinberg whispers how the alleles get older, bcftools counts columns, HaplotypeCaller rebuilds instead— local reassembly don't trust the pile, it walks the graph ahead. [Chorus] Pileup or assemble, pick your witness stand, prior meets the likelihood, posterior in hand, joint call the cohort, gVCF's the bridge, filter out the liars from the ridge to the ridge— QD and strand bias, allele balance don't lie, hard filters or VQSR, teach the model why. [Verse 2] (Female) DeepVariant reads the pileup like a photograph, convolutions scanning for the variant epitaph, Clair3 rides the long reads, context is the key, neural nets replacing the simple MLE, cohorts bank in gVCF, block by silent block, GenotypeGVCFs turns that bank into a stock, rare variants get rescued once the priors tighten up— but low QD's a red flag, don't drink from that cup. [Chorus] Pileup or assemble, pick your witness stand, prior meets the likelihood, posterior in hand, joint call the cohort, gVCF's the bridge, filter out the liars from the ridge to the ridge— QD and strand bias, allele balance don't lie, hard filters or VQSR, teach the model why. [Verse 3] (Male) Tumor next to normal, subtraction tells the story, Mutect2 hunts the somatic, low VAF, no glory, panel of normals scrubs the sequencer's scars, FFPE fakes a mutation, oxidation draws its bars, Strelka2 races clean, somatic and precise, purity and ploidy — the ASCAT roll of dice, subclones hide in fractions, point-oh-five to spare, cancer's a population evolving unaware. [Verse 4] (Female) Structural variants brawl in four kinds of evidence— split reads snap in half, discordant pairs make sense, Manta stitches seams, Delly counts the pairs, Sniffles rides long reads where cuteSV repairs, CNVkit normalizes GC, segments what we see, CBS carves the genome, Markov walks it free, ExpansionHunter counts the triplet, repeat by repeat, WhatsHap phases reads, trios make it complete. [Bridge] (Male) Left-align the indel, vt and bcftools norm it, (Female) same mutation, different clothes — decompose, transform it, (Both) GIAB's the mirror, hap.py grades the test, precision, recall, F1 — stratify the rest, homopolymers, segdups, the regions called "hard"— truth is not one number, it's context, it's a guard. [Chorus] Pileup or assemble, pick your witness stand, prior meets the likelihood, posterior in hand, joint call the cohort, gVCF's the bridge, filter out the liars from the ridge to the ridge— QD and strand bias, allele balance don't lie, hard filters or VQSR, teach the model why.
← Module 7 — Sequence Algorithms II: Assembly, k-mers & Sketching (Weeks 12–13) | Module 9 — Variant Interpretation & Annotation (Week 17) →