Skip to content
Tech Interview Prep

Top 100 Computational Biologist Practice MCQs

Every question on one page: open one, choose an answer, and see why at once. Signed in, your latest answer to each question counts, and you can answer again after 24 hours.

Curated: · Written: · Reviewed:

Reviewed 100Review pending 0

Answers to reviewed questions count toward this role and the overall leaderboard; your latest answer to each one is the one that counts.

100 questions

MCQ-84A candidate is implementing Needleman-Wunsch for global alignment of two sequences using a match score of +1, a mismatch score of -1, and a linear gap penalty of -2. The scoring table is filled cell by cell. Which filling and traceback procedure produces the optimal global alignment?
MCQ-3A protein aligner is configured with match +2, mismatch −3, gap open 11, gap extend 1. The pipeline it replaces used a linear gap penalty of −4 per residue. Two alignments of the same 72-residue query to a subject are proposed. Alignment X has 50 matches, 10 mismatches, and one 12-residue gap. Alignment Y has 50 matches, 10 mismatches, and twelve separate 1-residue gaps. Under the affine model, which alignment scores higher, and by how many points?
MCQ-2A 800-residue fusion protein is compared with a 180-residue domain from its partner. A colleague runs Needleman-Wunsch with default gap penalties and reports 88% identity. The alignment shows the 180-residue domain aligned to a 180-residue segment of the fusion, with the remaining 620 residues on the fusion side placed opposite gaps. What should you conclude about the 88% figure and do next?
MCQ-1A 510-residue query protein is searched with BLASTP against a 40-million-residue non-redundant protein database. Two hits pass the E-value cutoff: | Hit | Identity | Alignment length | E-value | Bit score | | --- | --- | --- | --- | --- | | A | 92% | 38 | 2e-11 | 70 | | B | 34% | 486 | 1e-96 | 350 | Before transferring a name, the team must decide what the search supports. Which conclusion is correct?
MCQ-4A team hunts remote homologs of a 300-residue enzyme: 400 queries, true hits averaging 22% identity to the query. Run 1 used BLOSUM80 and recovered 138 of the 400; run 2 used BLOSUM45 and recovered 200, same database and E-value cutoff. Which reading of the two runs is right, and what should ship?
MCQ-85A blastp search of a 380-residue query against Swiss-Prot at word size 3 with the two-hit heuristic on recovers 41 of 50 known remote homologs (20–30% identity) in 9 minutes. The nine missed homologs share no exact 3-mer with the query. Dropping from word size 3 to word size 2 multiplies the number of seed hits roughly twentyfold. The team needs all 50 within a 2-hour wall-clock limit on the search step. Which configuration should they run?
MCQ-5A team annotates a 12 Mb bacterial genome (1.2 × 10^7 residues). Their SOP calls a protein a homolog when a BLASTP hit against the local genome database has E < 1e-6. A 300-residue query hits a 57-bit alignment with E = 2.5 × 10^-8 against the local database. For the paper, the same alignment is re-searched against nr (2.1 × 10^9 residues). What should the team report for the nr search?
MCQ-9A 480-residue coiled-coil query (SEG flags 61% of it as low-complexity) is searched against a 12-million-residue protein database. The unmasked run returns 8,400 hits; none of the 26 known homologs in the team's labeled set ranks above 500th, and the top 200 are short coiled-coil fragments with E-values from 1e-4 to 1e-9. A hard-masked run returns 26 hits but keeps only 4 of the 26 labeled homologs. Which configuration should the team report from?
MCQ-6A team has a 350-aa protein and a 130 Mb draft genome with no gene annotation. They need to locate the gene encoding this protein, and the gene is expected to contain introns. Which search locates it?
MCQ-8A variant-classification pipeline stores each BLASTp hit as a free-text line, e.g. 'NP_000001, 99% identity, E=1e-45'. The pipeline logs the command line (blastp -query q.fa -db nr -outfmt 6) and the BLAST+ version (2.16.0); both are byte-identical for the March run and an August re-run. The August re-run returns 'NP_000001.2, 97% identity, E=1e-40'. nr is rebuilt periodically. Which artifact must the pipeline record so the August report can reproduce the March call?
MCQ-13You are assembling a 4.6 Mb bacterial genome from 2×150 bp Illumina reads at 12× mean depth. The genome carries many short interspersed repeats of 40-60 bp. Your assembler needs at least 5× k-mer coverage to build a connected graph. You are choosing between k=31, k=71 and k=111. Which k yields the longest contigs?
MCQ-12A bird genome is going to release. The k-mer estimate of its genome size is 2.4 Gb, but the assembly pipeline's config still carries 3.2 Gb, copied from a mammal reference. The report shows contig N50 = 19 Mb and NG50 = 8 Mb, and the release checklist requires NG50 computed against the estimated genome size. What does the gap between N50 and NG50 show here, and what should the team do?
MCQ-14A diploid genome carries a 40 kb tandem repeat array at 11 copies per haplotype. The lab has 30x of 8 kb raw nanopore reads (Q10) and 20x of 45 kb HiFi reads (Q30). An assembly built from the 8 kb raw reads, then polished with short reads to Q40, reports 3 copies of the array. Which action corrects the copy number?
MCQ-15A diploid mammal is assembled with hifiasm from 30× PacBio HiFi; the read-depth histogram shows a diploid peak at 30× and a haploid peak at 15×. purge_dups (default settings) removes 180 Mb of contigs, including a 210 kb contig, ctg88. ctg88 has mean depth 15.2× and aligns to retained ctg31 over 94% of its length at 99.6% identity; ctg31 has mean depth 30.1×. Trio phasing places ctg31 intact on both parental haplotypes and ctg88 on the paternal haplotype only. What does this evidence establish about ctg88?
MCQ-16A diploid plant genome is 1.2 Gb. The assembly is 1.35 Gb with contig N50 2.1 Mb. BUSCO on embryophyta_odb10 reports C:97.2% (S:95.1%, D:2.1%), F:0.8%, M:2.0%. Merqury k-mer completeness is 85.0%. The funded trait locus lies in a 180 Mb repeat-rich pericentromeric block and is absent from the assembly. Which conclusion is best supported?
MCQ-11A 2.71 Gb mammalian genome is assembled twice from the same reads. Assembly A: N50 38.2 Mb, span 2.84 Gb, BUSCO complete 91.0%, 14 misassemblies, 130 Mb of contaminant sequence. Assembly B: N50 18.6 Mb, span 2.71 Gb, BUSCO complete 96.1%, 0 misassemblies, 0.4% contaminant. The team plans to release A because its N50 is twice B's. Which release decision is defensible?
MCQ-19An assembly of a 4.6 Mb bacterial chromosome from 30× HiFi reads yields a single 4.62 Mb contig whose ends share a 12 kb segment that is 99.9% identical to an IS element occurring eight times in the genome. Reads spanning the join: 0. Coverage at the join: 0.4× against a 32× genome mean. What is the defensible call?
MCQ-18A 3.1 Gb draft assembly of a wild accession (3% divergent from the only available reference) is polished for three rounds with short-read alignments. QV rises 28 → 42, BUSCO stays at 96.1%, misassemblies stay at 14, and discordant read pairs rise from 4.1e4 to 5.8e4. The 14 misassemblies are joins where the two contigs have no support from paired reads. What is the right next step?
MCQ-20A 30 Mb GC-poor fungal genome is sequenced with 2×150 bp Illumina reads; the run yields 1.2 Gb of bases, reported as 40× nominal coverage. After mapping, 55% of bases align to the 55 kb mitochondrial contig, 30% to nuclear unique sequence, and 15% to repeats or nothing. The assembly is missing 38% of BUSCO genes. Which conclusion about the coverage does the data support?
MCQ-86A short-read assembler (k = 31) reports a single 2 kb contig for a tandem array in a bacterial genome. Genome-wide unique k-mer depth is 40×; the k-mers inside the array's 2 kb unit each sit at 440×. HiFi reads span the array end to end as one contiguous tandem block. The annotation pipeline then reports one copy of a toxin gene that lies inside the unit. What should the pipeline conclude?
MCQ-87A team estimates genome size from a k-mer spectrum of a 3.0 Gb haploid genome sequenced at 30×, using k=21 and counting every k-mer on both strands. The pipeline reports 5.9e9 unique k-mers, calls the genome 5.9 Gb, and flags the library as undersequenced. What should the analyst do?
MCQ-99An overlap graph is built from 30x PacBio HiFi reads with a 1,000 bp minimum overlap at 97% identity. The karyotype and the Hi-C map (4 diagonal blocks) both give 4 chromosomes for the 160 Mb genome, but the graph resolves into 3 connected components: the largest holds two Hi-C blocks that sit on different chromosomes and share a 2.5 kb satellite repeat at 99% identity between copies (about 900 copies, present only on those two chromosomes). What does the component structure support?
MCQ-10A variant-calling pipeline aligns each 3.1 Gb genome to the reference with banded Smith-Waterman at band width k = 32 bp; a genome takes 4 minutes and the cluster of 64 cores processes 500 genomes/day. In one sample, a 1,200 bp homozygous deletion in an exon comes back as 40 clustered SNPs plus a frameshift, and the alignment path runs along the band edge across that exon. Unbanded alignment of the same 5 kb region recovers one 1,200 bp gap and no SNPs. Unbanded whole-genome alignment takes 6 hours per genome. What should the pipeline do?
MCQ-21A joint callset merges 14,812 sites that an rsID-based intersection calls shared between GRCh37 and GRCh38. The pre-merge report shows 12,771 sites have REF matching the GRCh38 FASTA, 2,041 sites have REF mismatch, and all 25 contig lengths match. If the pipeline keeps all 14,812 records, how many records will have a REF base that does not match the GRCh38 FASTA?
MCQ-22A pipeline converts a VCF to BED for interval overlap against ClinVar. The BED consumer expects 0-based half-open intervals, the callset contains SNPs, insertions and multi-base deletions, and the caller emits INFO/END only for symbolic alleles. Which conversion rule should the layer implement?
MCQ-24A pipeline splits multiallelic records into biallelic records. For CHROM=1 POS=100 REF=A ALT=C,T, sample S1 has GT=0/2, AD=12,3,40. After the split, the T record reads GT=0/1, AD=12,3,40, and a reviewer flags it. What should the T record's GT and AD be?
MCQ-25A germline pipeline is validating two callers on the same 30x sample against the same GRCh38 FASTA. In a 10 bp poly-A run at chr7:1001-1010, caller A emits one record: POS=1001 REF=AAA ALT=A. Caller B emits two records phased on the same haplotype: POS=1001 REF=AA ALT=A and POS=1004 REF=AA ALT=A. A merge script that compares POS and REF/ALT as strings reports the locus as discordant, blocking the merge. What should the team conclude and do?
MCQ-23A germline pipeline filters a callset with GATK 4.4 VariantFiltration (`--filter-expression "QD < 2.0" --filter-name LowQD`), then merges it with an unfiltered gVCF-derived callset. The merged VCF has 4,180,000 records: 3,940,000 PASS, 12,400 LowQD, and 227,600 '.' from the unfiltered callset. The lab's SOP requires every variant in the clinical report to have passed the QD filter, and the report script selects `FILTER == "PASS"`. A colleague proposes adding the 227,600 '.' records to the report because "the filter didn't flag them." Which handling is correct?
MCQ-88A WES pipeline annotates variants against an exon BED. For a clinically relevant exon the BED line is `chr2 4801 4900 EXON3` (0-based half-open). The annotation code keeps a variant when `start <= POS < end`, reading POS straight from the VCF. A pathogenic splice variant at POS 4900, the last base of that exon, is reported as intronic, while a benign variant at POS 4801, one base upstream of the exon, is reported as exonic. Which change fixes the cause?
MCQ-89A Python 3.12 pipeline reads a 40 GB FASTQ file in 100 KiB binary chunks, splits each chunk on `b'\n'`, and processes lines in groups of four. It runs in ~30 MB RSS. Downstream, 0.3% of reads have sequence length 151 and quality length 150. Which change addresses the cause?
MCQ-26A trio is joint-genotyped for a recessive disorder: each sample was run through GATK 4 HaplotypeCaller in -ERC GVCF mode, then GenomicsDBImport and GenotypeGVCFs produced one sites-only VCF. The proband carries a heterozygous frameshift in GENE3 inherited from the mother. The father has no GENE3 variant row, and his gVCF holds no <NON_REF> block over GENE3 exons 3-5; the BED interval list used for his HaplotypeCaller run omitted those exons. The mother's gVCF has <NON_REF> blocks with DP around 30 across GENE3. What does the father's result mean for the compound-het search?
MCQ-27A 2021 study produced 8,000 CRAM 3.0 files. The archive kept a FASTA labelled GRCh38 from a mirror. A restore drill decodes 800 randomly chosen files with samtools; 6 fail with a reference MD5 mismatch on chr2, and 794 decode. The mirror's chr2 differs from the reference used at write time by 3 bases. What should the team do?
MCQ-28A WGS cohort is being released. One candidate pathogenic het is supported by 18 of 30 reads, all with MAPQ 0 and base quality ≥ 35; every other called site in the callset has MAPQ ≥ 40. The locus lies in a segmental duplication, and a paralog 8 Mb away is 99% identical and carries the alternate base at that position. What should the analyst conclude about this call?
MCQ-29A tumor of 30% purity, copy-neutral at the locus in question, and its matched normal are each sequenced to 80x. A somatic SNV at that locus is heterozygous in the tumor cells: one mutant copy, one wild-type. An analyst runs GATK HaplotypeCaller in its default diploid germline mode on the tumor BAM alone and delivers the unfiltered output as the somatic mutation list. Which statement about this locus in that callset is correct?
MCQ-30A clinical lab is validating a SNP caller against GIAB HG002 v4.2.1 before using it for MHC typing. The caller emits 3,200,000 SNP calls: 3,150,000 inside the GIAB confident BED and 50,000 outside it. Inside the BED, 3,120,000 of the calls match the truth set, 30,000 do not, and 60,000 truth SNPs are missed; the truth set holds 3,180,000 SNPs there. Which F1 belongs in the validation report?
MCQ-31A lab sequences 12 WGS samples (mean depth 32x) with a new library prep kit and runs one NA12878 control in the same batch. GATK VQSR with the bundled 1000 Genomes training and truth resources at the 99.9 SNP tranche yields 4.1M SNPs and a clean bimodal VQSLOD histogram. Against the NA12878 truth set, VQSR recall is 0.58; the lab's predeclared hard filters give 0.91 on the same raw callset. What does this comparison indicate?
MCQ-32A pipeline's converter turns each VCF breakend pair into rows of its SV table. Its readers take one row as one event, merge two rows into a single junction only when each names the other in MATEID, list a merged junction under both of its rows' chromosomes in the per-chromosome counts, and the tumor board prints any row whose end minus start exceeds 1 Mb as a deletion, whatever its SVTYPE says. For a hypothetical balanced t(1;2) the caller emits chr1 1234567 with ALT N[chr2:45678901[ and MATEID=bnd_B, plus chr2 45678901 with ALT ]chr1:1234567]N and MATEID=bnd_A; the CNV caller reports no copy-number change at either locus. The converter reads only the first record's CHROM and POS, writes one row chr1 1234567-45678901, and drops the second record, the MATEID and the brackets, so the board prints a 44.4 Mb chr1 deletion, the chr1 count takes it as a junction and chr2 shows none. Only the converter can change. Which change fixes the report?
MCQ-81A germline pipeline stores per-sample mean depth as a float and gates on `mean_depth >= 30`. A refactor rounds the stored value to the nearest integer before the gate. In a 240-sample cohort, 14 samples have a stored mean of 29.6× and 71% of bases at ≥20×; the other 226 have ≥30× and 94% of bases at ≥20×. Variant recall against a truth set is 0.88 in the 14 and 0.97 in the 226. The callset is due to ship. Which change should the review make before it ships?
MCQ-72A 30-sample germline pipeline is rerun three weeks after its first run. Same script, same FASTQ files, same host. Week 1: 99.1% of reads aligned, 4,210 indel calls. Week 2: 97.0% aligned, 4,390 indel calls. The config reads `image = bwa:latest`. The week-1 run log records image ID `sha256:9f2c…`; `docker inspect bwa:latest` today returns `sha256:41ab…`. A control run with the image pinned to `sha256:9f2c…` reproduces 99.1% and 4,210 indels on the same FASTQs and reference index. Which conclusion does the evidence support?
MCQ-73A tumor-normal pair is processed with a somatic caller. The sample sheet lists tumor T and normal N, and barcodes match the sheet. The caller reports 4,100 somatic SNVs. Of these, 3,900 are common germline SNPs with VAF near 50% in both T and N. A 50-marker SNP fingerprint shows T and N share only 12 genotypes; each sample's fingerprint matches its own tube, but the two tubes do not match each other. What should the team conclude and do before releasing the callset?
MCQ-7A team annotates a 70%-complete draft assembly of species A against the complete genome of species B. Reciprocal best hits (RBH) yield 6,400 candidate 1:1 orthologs. For one pair, A-g1 ↔ B-g1, a gene tree built from the two sequences plus their two closest paralogs places A-g1 sister to A-g2, a tandem duplicate in A, with B-g1 outside that clade. What should the team do with this pair?
MCQ-17A stool metagenome is sequenced with PacBio HiFi (15 Gb) and assembled with hifiasm; median contig depth is 40x and median read length is 12 kb. Three extraction blanks, each sequenced to 1.5 Gb, were assembled the same way. The pipeline's only screen deletes any contig with a 99%-or-better nucleotide hit to a non-host reference. Two contigs hit the same K. pneumoniae reference sequence: ctg_112 (48 kb, 99.4% identity over 44 kb, GC 57.1%, depth 41x, absent from all three blanks, and spanned by 14 HiFi reads that continue into ctg_003, which carries a 16S rRNA sequence 99.7% identical to K. pneumoniae) and ctg_088 (96 kb, 99.1% identity over 93 kb, GC 57.2%, depth 3.1x, hit by 11 reads in two of the three blanks, no junction-spanning reads). Which calls do these data support?
MCQ-33A team sequences RNA from 2 treated and 2 control mice. Each mouse's library is split across 3 sequencing lanes, and the analysis treats all 6 libraries per group as independent replicates. DESeq2 reports 1,842 genes at FDR 0.05. Which statement about the result is correct?
MCQ-34A pilot RNA-seq study has 16 libraries: 8 treated, 8 control. A reagent shortage forced all 8 treated libraries to be prepped on day 1 and all 8 controls on day 2. The analyst fits a model with design ~ condition + prep_day. Which statement about the model matrix and the resulting analysis is correct?
MCQ-35Twelve Salmon-quantified libraries (6 control, 6 treated) are imported at gene level with tximport. The team builds the DESeq2 matrix from TPM values rounded to the nearest integer. DESeq2 returns 1,842 genes at FDR < 0.05, and its size factors span 0.98–1.03. The same libraries imported as estimated counts (countsFromAbundance = 'no') return 213 genes, with size factors spanning 0.71–1.42. Which explanation fits these numbers?
MCQ-36A bulk RNA-seq experiment compares 4 drug-treated with 4 control liver samples, each sequenced to the same depth (~40M reads). One gene, Cyp3a11, is strongly induced: it holds 2% of aligned reads in controls and 35% in treated. The analyst normalizes with CPM (counts per million, total counts) and reports 1,100 genes down at |log2FC| > 1. For a gene whose true expression is identical in both groups, what apparent fold change (treated/control) does CPM produce?
MCQ-37A facility sequences 24 libraries with a dUTP protocol and the sample sheet records them as forward-stranded. For one library, featureCounts on the same BAM assigns 41% of fragments with -s 1 and 92% with -s 2, and RSeQC infer_experiment.py reports 0.93 of fragments in the 2+-,2-+ class. What should the analyst do for the cohort?
MCQ-38A 24-sample RNA-seq study tests 12,000 genes with DESeq2 and Benjamini-Hochberg across the whole table: 500 genes pass at padj < 0.05 and 40 at padj < 0.001. The abstract draft says 'each of the 500 genes has a 95% probability of being a true effect, and the top 40 are at least 99.9% likely to be real.' The reviewer asks how many false positives the team expects in these lists and how the per-gene percentages are justified. Which reply is defensible?
MCQ-39A team reports an interferon-stimulated gene (ISG) score up 6.0-fold in bulk RNA-seq of 20 treated vs 20 untreated tumors, and the score is dominated by CD8 T-cell expression. Deconvolution of the same samples puts the CD8 T-cell fraction at 10% in untreated and 30% in treated tumors. Matched single-cell RNA-seq from the same tumors measures a 1.4-fold ISG induction within CD8 T cells. Which conclusion does the bulk data support?
MCQ-41Eight patients are biopsied before and after a drug, giving 16 libraries, one per sample. limma-voom is run on log2-CPM with two design formulas: ~ condition, and ~ patient + condition. For gene G the treatment contrast is 40 log2 units in both fits. The SD of the eight within-patient differences is 21, and the SD of the 16 libraries within each condition is 60. Which account of the two fits for gene G is correct?
MCQ-42A DESeq2 comparison of 3 control against 3 treated RNA-seq libraries returns for gene X: mean normalized counts of 4.0 and 0.3, an unshrunken MLE log2 fold change of 3.7, a value of 1.1 after lfcShrink with apeglm, and a Wald test padj of 0.62. The team's figure ranks genes by effect size and marks those with padj below 0.05. What should gene X's row in that figure report?
MCQ-43A height GWAS in 80,000 adults recruited along a north-south cline reports 4,200 loci at p<5e-8 with no ancestry covariates. The top hits' effect sizes correlate with PC1 (r=0.6), PC1 correlates with recruitment latitude (r=0.8), and genomic inflation is λ=1.35. What does this pattern indicate, and what should the team do?
MCQ-44A GWAS of 25,000 samples includes 3,000 sibling pairs. Linear regression assuming unrelated individuals returns lambda_GC = 1.09 and 38 genome-wide significant loci. A mixed-model refit gives lambda_GC = 1.01 and 12 loci. The funder requires all 25,000 samples in the final analysis. Which action is right?
MCQ-45A GWAS meta-analysis of 50,000 cases and 50,000 controls tests 1,000,000 imputed common SNPs (MAF > 1%) and 20,000 rare-variant gene-based burden tests (MAF < 1%). The predeclared analysis plan sets a common-variant threshold of 5e-8 and a burden threshold of 2.5e-6 (Bonferroni over 20,000 genes). Results: rs123 p=3.2e-8; gene ABC burden p=1.1e-6; gene DEF burden p=4.0e-6. Which set should the paper report as genome-wide significant?
MCQ-46A GWAS of 12,000 cases and 15,000 controls in an African-ancestry cohort finds lead SNP rs123 (P = 3×10^-12) in an intron of gene X. Fine-mapping with an ancestry-matched LD reference gives a 95% credible set of 18 variants; one is an enhancer 80 kb away. In liver eQTL data, that enhancer variant is associated with gene Y (P = 2×10^-8), while rs123 is not. The team must decide what to report as the likely causal mechanism. Which action is best?
MCQ-47Nine cohorts contribute to a fixed-effect meta-analysis of a single A/T SNP. Each cohort's |beta| is at least 0.12, yet the pooled beta is 0.01. What is the correct diagnosis and fix?
MCQ-48A case-control GWAS (4,000 cases, 4,000 controls) tests a recessive model on rare coding variants. Hard-calling at INFO ≥ 0.3 returns 47 genome-wide significant loci; re-running the identical model on expected genotype dosages returns 6. All 41 dropped loci have MAF below 1% and INFO between 0.3 and 0.6. Which conclusion do these numbers support?
MCQ-50A 4,000-case / 4,000-control GWAS applies one HWE chi-square test (1 df) per variant to all 8,000 samples pooled and drops variants below 1e-6. For rs1234 in the MHC, controls are AA 1,670 / AB 1,850 / BB 480 (chi-square 0.88, p = 0.35); cases are AA 2,100 / AB 1,400 / BB 500 (chi-square 111, p < 1e-25); pooled chi-square is 45, p = 2e-11. Case and control allele frequencies are 0.70 and 0.649 (allelic OR 1.26, p = 5e-12). The pipeline drops rs1234. What should the QC step do with it?
MCQ-51A GWAS of 40,000 cases and 60,000 controls has lambda_GC = 1.15. The team runs LDSC twice. With ancestry-matched LD scores: intercept = 1.16 (SE 0.01), slope = 0.02 (SE 0.03), mean chi-square = 1.18. With LD scores from a different ancestry: intercept = 1.02 (SE 0.02), slope = 0.15 (SE 0.04). Which conclusion is defensible?
MCQ-52A 240-isolate bacterial cohort is ready for release. The core-genome alignment has 1,842 SNP sites. IQ-TREE run A (seed 1) and run B (seed 2) both place isolate C-04, a known lab contaminant, as the sister to the P-17 clinical clade with 100% UFBoot support. Removing a single 312-bp chimeric region from the C-04 assembly moves C-04 to a basal position and removes the P-17 sister relationship; rerunning both seeds now gives the same basal topology. The supplement currently contains only the run A tree and a FastQC plot. What should the team release?
MCQ-53A 30-taxon amino-acid alignment has two taxa, X and Y, with branch lengths 2.5× the median; both are AT-rich (68% and 66% AT), while the short-branch taxon Z is 45% AT. GTR+G recovers (X,Y) as sisters at 95% bootstrap. A χ² test rejects compositional homogeneity across the 30 taxa (p < 0.001), and after removing the 500 most compositionally biased sites, Y groups with Z at 88% bootstrap. Which conclusion does this evidence support?
MCQ-54A 1,200-taxon phylogenomic study wants to name a new subclade, and the journal requires at least 70% bootstrap support computed by resampling characters. The authors' IQ-TREE 2 run used `-bb 1000` and reports the clade at 96; a reviewer reruns the same alignment and model with `-b 1000` and gets 88. What should the authors do?
MCQ-55A 2,000-gene phylogenomic dataset for four taxa (P1, P2, P3, outgroup O) yields a concatenated tree ((P1,P2),P3) with 100% bootstrap support. Gene-tree counts are 1,240 for ((P1,P2),P3), 600 for ((P1,P3),P2) and 160 for ((P2,P3),P1). An ABBA-BABA test gives D = 0.31 with Z = 6.2. Which conclusion is supported?
MCQ-56You align 12 near-full-length HIV-1 genomes (9,000 bp) from a recent outbreak. A maximum-likelihood tree on the concatenated alignment places sample E12 on a long branch between subtypes B and C with 100% bootstrap support. When you split the alignment, E12 clusters with subtype B in gag (positions 1–1500, 98% bootstrap) and with subtype C in pol (positions 2500–4500, 97% bootstrap). What should you conclude?
MCQ-57A 24-taxon phylogeny is built from a 1,200-column alignment. On the full alignment the two parasite genera are monophyletic with UFBoot 58%. A Gblocks run with default settings removes 190 columns (16%), and on the trimmed alignment the genera are no longer monophyletic: one genus now groups with the host outgroup at UFBoot 96%. The Gblocks settings were chosen after the full-alignment tree was seen. Which action does the evidence support?
MCQ-58AlphaFold2 models a two-domain protein: domain A (residues 1–180) and domain B (residues 220–400) joined by a 40-residue linker. Both domains score pLDDT 93. The PAE matrix gives 22 Å between every A–B residue pair. The team wants to design a disulfide that pins A against B. What should they do?
MCQ-59A team is about to run a virtual screen against a kinase domain. Their only structure is an AlphaFold2 monomer model (no ligand, no partner). In an internal retrospective control (hypothetical figures), they docked 200 known actives from a holo crystal structure into the AF2 model; none scored in the top 1%. Superposing the AF2 pocket onto the holo crystal gives 4.2 Å RMSD over the gatekeeper residues, while the AF2 pLDDT in that pocket is 92. They have one week and no experimental holo structure for this kinase. Which action is defensible?
MCQ-60A team is modeling the dimer interface of an enzyme for a structure-based design campaign. The deposited mmCIF has a single protein chain, A, in the asymmetric unit. The `_pdbx_struct_assembly_gen` record specifies assembly 1 as two copies of A generated by symmetry operators, with a total buried surface of 1,240 Ų across the A–A' interface. A colleague runs an interface calculation on the deposited coordinates alone and reports "no interface found: the protein is monomeric." Which assessment is correct?
MCQ-61A team models a 400-residue protein against its crystal structure. All-atom RMSD over all 400 residues is 8.6 Å; the 350-residue core superimposes at 1.9 Å, while the 50-residue C-terminal tail is modeled as extended and deviates by about 20 Å. TM-score is 0.83. What should the report conclude about the fold?
MCQ-62An enzyme-mechanism paper is under review. The catalytic domain was solved at 2.6 Å with a substrate analog bound: Ser-102 points into the active site, its hydroxyl 2.8 Å from the analog's reactive carbon, with a side-chain real-space correlation of 0.88. An AlphaFold2 model of the same sequence, built without the analog, places Ser-102 pointing into solvent, pLDDT 95, with a 4.0 Å PAE between the two domains. S102A leaves 0.5% of wild-type activity. A referee asks which rotamer the mechanism figure should show. Which decision is correct?
MCQ-63A 3.2 A cryo-EM map of a 312-residue membrane protein refines to 97.8% Ramachandran favored with 0.4% outliers, and the residues of helix 112-124 are among the favored set. Main-chain density along that helix is strong and continuous, but its local map CC is 0.38 while the rest of the same chain sits at 0.78. Which next step does this evidence support?
MCQ-92A 2.1 Å kinase–inhibitor complex refines to Rwork/Rfree = 0.19/0.23 with a whole-model MolProbity clashscore of 4.2 over 8,400 atoms. The 24 residues lining the ATP pocket hold 190 atoms and 8 clashes, and the inhibitor pose is the basis of a figure. What does the validation data support?
MCQ-64A team trains a cell-type classifier on 200,000 cells from 48 donors. Five-fold cross-validation that splits cells at random reports AUC 0.97. The same pipeline, evaluated on 8 donors held out entirely, scores AUC 0.62. The model's top-weighted gene is near-constant within a donor but varies 8-fold between donors. Which explanation accounts for the gap?
MCQ-65A team trains a variant-effect predictor on 120,000 missense variants from 2,000 genes. A random 80/20 split by variant gives test AUC 0.94; a 30%-identity cluster split by gene family gives 0.71. The model will be deployed to score new variants in the same 2,000 genes, submitted by clinicians. Which number best predicts deployment performance?
MCQ-66A team tunes a gradient-boosted model on 2,000 labeled rows. A random search scores 120 hyperparameter configurations on the same 5-fold split of those rows and quotes the best configuration's fold-mean AUC, 0.91. A frozen copy then scores 0.84 on an untouched 500-row prospective set, and 0.85 on a second untouched 500-row set drawn at random from the same period. Which account of the gap and which number should the team report?
MCQ-67A sepsis-alert model is up for deployment. On a held-out set of 100,000 patients drawn from the same 1%-prevalence population it will serve, AUC is 0.90; at the reviewed cutoff, sensitivity is 90% and specificity is 90%. Which conclusion is defensible from these numbers?
MCQ-68A variant-effect CNN is trained on 1000 Genomes variants from chromosomes 1–7 and 9–22; chr8 is held out. A random 80/20 SNP split scores AUC 0.94 (calibration slope 0.98); the chr8 holdout scores AUC 0.71 (slope 0.61). The model card needs one generalization estimate for a new cohort. Which entry is correct?
MCQ-69A germline SNP caller was trained only on NovaSeq 6000 2x150 data. Its NovaSeq validation on GIAB HG002 at 30x gave SNP F1 0.9985 and 300 false-positive SNPs per genome. The release gate for any chemistry is SNP F1 >= 0.995 and false-positive SNPs <= 500 per genome on GIAB HG002 at 30x. The team now runs the same caller on a new 2x150 chemistry and measures: | Calibration | SNP F1 | FP per genome | | --- | --- | --- | | none | 0.982 | 5,200 | | homopolymer filter only | 0.991 | 600 | | recalibrated on 10x GIAB | 0.994 | 720 | | recalibrated on 20x GIAB | 0.996 | 410 | All runs are evaluated on GIAB HG002 at 30x. The new chemistry's Q30 is 92 percent, and the release notes have no per-chemistry GIAB table. Which release decision follows from these measurements?
MCQ-70A rare-disease genomics lab runs a variant-prioritization model inside its research pipeline. Model card v4 records: - intended use: research prioritization of candidate variants in adult probands - excluded use: using the model's flags to select variants for clinical return of results - validation: 4,100 adult probands (78% EUR, 9% AFR), held-out AUC 0.93 - operating point: score >= 0.62; false-discovery rate 5.4% overall, 21% in the AFR subgroup (371 samples); artifact hash 9f3c, dataset build GRCh38-v12 The team proposes one change: wire the same model, unchanged at 0.62, into the return-of-results queue so flagged variants reach the ordering clinician's report for an adult research cohort of 1,200 probands that is 34% AFR. They ask the review board to approve the change. Which action should the board take?
MCQ-49A hospital genomics group trained a coronary artery disease PRS on a 400,000-participant European-ancestry GWAS; in held-out Europeans the AUC is 0.68. They apply the same weights, unchanged, to 9,000 African-ancestry patients. In that cohort the AUC is 0.63, the calibration slope is 0.52, and among patients in the top decile of the score the observed 5-year event rate is 4.1% against a predicted 10.3%. The plan is to offer preventive statin therapy to everyone above the 90th percentile of the score. What should the group do?
MCQ-40A pipeline joins a counts matrix to an annotation table on HGNC symbol. The counts were quantified against Ensembl release 113; the annotation table was built from release 110. After the join, 37 genes in the counts matrix have no annotation row, and the symbol C1orf106 matches two different Ensembl gene IDs. The team needs the join to reproduce the exact gene model used for quantification and to fail loudly if a symbol is reused. Which change should they make?
MCQ-71A 96-sample RNA-seq run adds 55 new samples to the sample sheet. The other 41 keep the same sample IDs and FASTQ paths as run 1, but a collaborator re-trimmed those 41 FASTQ files in place, overwriting the originals. The pipeline keys its per-sample cache on sample ID plus FASTQ path, so run 2 reports 41 cache hits and runs the 55 new samples, then exits 0 in 11 minutes. QC on the 41 cached BAMs reproduces the run-1 plots exactly. What should the team do about the 41 hit samples?
MCQ-74A genomics lab publishes 5,000 human whole-genome sequences to a public S3 bucket, arguing that FAIR requires open access. Within a month, a researcher matches eight participants to a genealogy site using Y-STR haplotypes read straight from the public files. The IRB orders remediation, and the funder still requires the cohort to be FAIR. Which action satisfies both?
MCQ-75Your lab's pipeline writes FASTQ to `/scratch` and logs to `/work/run_9f3a/logs`. `/work/run_9f3a` is group-readable by 12 lab members; `/scratch` is world-readable and indexed by `mlocate`, which rebuilds nightly at 02:00. A run used `/scratch/JaneDoe_tumor_R1.fq`; `sample_map.tsv` in `/work/run_9f3a` maps six pseudonyms to patient names and MRNs; `logs/align.stdout` contains the original `/scratch/JaneDoe_tumor_R1.fq` path. `locate JaneDoe` returns six paths. The IRB requires no patient names or MRNs in shared scratch, shared work directories, or logs; the mapping may live only in a 600 file outside shared spaces. The next run is at 12:00, and the cluster admin can purge specific paths from the mlocate index within the hour if ticketed. Which action is defensible?
MCQ-76A CNV-calling pipeline runs 96 tumor whole-genome samples under a fixed 32 GB memory limit. All 88 samples with under 3 GB of input finish; all 8 samples above 3 GB die with OOM. Among the 88 that finished, 1 shows whole-genome duplication (WGD). Re-run on a 128 GB node, 7 of the 8 OOM samples show WGD. The team is about to publish a "CNV landscape" reporting WGD prevalence from the 88-sample callset. What should they conclude?
MCQ-77A phosphoproteomics pipeline joins a paper's site table onto UniProt by accession, stripping any isoform suffix before the join. The paper reports a phosphosite at Ser387 of the isoform-2 record of a protein (429 residues). The pipeline joins it to the canonical record (393 residues), finds 387 ≤ 393, and reports the site as present and conserved across 40 species. What should the reviewer require before the mapping is accepted?
MCQ-78A team runs GO term enrichment on 400 genes from a differential-expression experiment and gets "protein kinase activity" at p=1e-12. Checking the annotations, 320 of the 400 genes carry only IEA (electronic) evidence, and the IEA annotations for most of them come from automated BLAST transfers onto the same kinase family the analyst had already named in the hypothesis. Restricting the same test to genes with IDA or IMP (experimental) annotations leaves 18 genes and p=0.4. What should the analyst report?
MCQ-79A team builds an eQTL catalog by overlaying RNA-seq (aligned to GRCh38, quantified against the Ensembl release 110 GTF) with variants (annotated against the Ensembl release 113 GTF). Both releases are built on GRCh38. A reviewer finds that 740 of the 12,000 reported eQTLs fall in exons that exist only in release 113. The team proposes to fix this by lifting the VCF coordinates from release 113 to release 110 with an Ensembl chain file. What does that lift accomplish?
MCQ-80`de` has 18,402 rows, one per gene, with a unique `ensembl_id` and a unique `symbol`. `ann` has 19,110 rows with unique `ensembl_id`; 708 of its symbols appear twice, each with a different Ensembl ID and biotype. An analyst runs `de.merge(ann, on='symbol')` and gets 19,110 rows. Which merge returns 18,402 rows, one annotation row per DE row, with each DE row matched to the annotation entry for the same gene?
MCQ-82A dashboard feeding a rare-disease power calculation counts carriers with this query: SELECT COUNT(*) FROM samples s JOIN variants v ON v.sample_id = s.sample_id WHERE v.chrom = 'chr7'; The samples table holds 1,200 samples. A quarter of them carry a chr7 variant, and each of those carries 8 chr7 variant rows; the rest carry none. The dashboard reports 2,400, and the analyst is about to plug that figure in as the number of chr7 carriers. Which number belongs in the power calculation?
MCQ-83A nightly job must write exactly one row per gene into `gene_primary(gene_id, transcript_id)`, choosing the transcript with the highest TPM and making the same choice on every run. The source is `expression(gene_id text, transcript_id text, tpm numeric, cds_length int)` on PostgreSQL 16, and the current statement is: ```sql INSERT INTO gene_primary SELECT DISTINCT ON (gene_id) gene_id, transcript_id FROM expression ORDER BY gene_id, tpm DESC; ``` Between runs, about 400 of 61,000 genes get a different transcript. Gene ENSG00000123456 has three transcripts (ENST00000311111, ENST00000322222, ENST00000333333) all at tpm = 12.0 and cds_length = 1842. Which change makes the job deterministic while keeping one row per gene?
MCQ-93A germline pipeline reads a 96-row samplesheet into `samples`, a channel of `[id, bam]` tuples, and builds `intervals` with `Channel.fromPath('intervals/*.interval_list')`, which matches two files. The author wires `HaplotypeCaller(samples.combine(intervals))`. The process declares `tuple val(id), path(bam), path(intervals)` and publishes `"${id}.g.vcf.gz"` to `results/`. The run finishes green. What did it produce?
MCQ-95A 12-sample pipeline loads each FASTA record into a mutable dataclass `Read(id, seq)`. QC writes `sha256(seq)` into a manifest; the trimmer then runs `read.seq = read.seq[:140]` on the same object. A reviewer re-hashes the reads in the BAM: 9 samples match the manifest and 3 do not, and the 3 are exactly the samples whose reads were longer than 140 bp. The team needs the manifest to let a reviewer verify both the sequence QC saw and the sequence the aligner used. Which change achieves that?
MCQ-98An aligner emits 5,000,000 HSPs per sample from parallel shards, so their arrival order varies between runs. The pipeline keeps the top 100 by bit score, and the worker container is capped at 512 MB. The current code sorts the full HSP array before slicing, which OOM-kills the worker. A first fix stopped the OOM by keeping only the best 100 in memory while streaming the HSPs, but two runs over the same input now retain different hits: 37 of the 100 differ, and every differing pair has an identical bit score. What should the team do?
MCQ-96A coverage QC table has one row per sample_id and contig_id: coverage(sample_id, contig_id, mean_cov). Sample S1 has 24 contigs: 23 at 31.3× and one contig, chrM, at 0×. The current query is: SELECT sample_id, AVG(mean_cov) AS sample_mean FROM coverage GROUP BY sample_id HAVING AVG(mean_cov) < 10; It reports S1 as 30.0× and passes, so the 0× chrM is never listed. The QC must return each sample's mean coverage and every contig with mean_cov = 0 for samples that either have a 0× contig or a sample mean below 10. Which single query change fixes the cause?
MCQ-97A coverage pipeline merges CDS intervals before counting covered bases. Intervals are 0-based, half-open, and the merge rule is: sort by start, then sweep, extending the current interval's end when the next interval's start is strictly less than the current end. The input on one gene is [100,500), [200,300), [350,400), [500,600), [700,800). Which merged output does the sweep produce?
MCQ-94An RNA-seq study of 24 tumours and 24 matched normals is annotated against 19,600 protein-coding genes. Genes with TPM >= 1 in at least 12 samples survive filtering, leaving 11,800, and 400 genes reach FDR < 0.05 with |log2FC| >= 1. Over-representation analysis is then run with the tool's default background of all 19,600 genes; GO:0007049 (cell cycle) has 2,000 annotated members, 1,900 of which passed the filter, and 96 cell-cycle genes are in the DE list. Which result should the paper report?
MCQ-90A team runs DESeq2 on 20,000 genes and reports 40 genes at padj < 0.1. To find more, they keep the 4,000 genes with |log2 fold change| > 0.5, re-run Benjamini–Hochberg on just those 4,000 p-values, and report 180 genes at padj < 0.1. Which statement about the 180-gene list is correct?
MCQ-91A two-sample Mendelian randomization study estimates the effect of LDL cholesterol on coronary artery disease. It uses 43 LDL-C instruments (all p < 5×10⁻⁸, mean F = 68), with GLGC as the exposure GWAS and CARDIoGRAMplusC4D as the outcome GWAS. Results: IVW OR 1.68 per 1-SD LDL-C (95% CI 1.52–1.86); MR-Egger intercept 0.021 (SE 0.007); MR-Egger slope OR 1.19 (0.92–1.54); weighted median OR 1.58 (1.40–1.79); Cochran's Q p = 0.004. One instrument is also associated with BMI (p = 2×10⁻¹¹). Which conclusion is defensible?
MCQ-100A GWAS tests 1.1 million common SNPs (MAF > 1%) against 20 correlated cardiometabolic traits in 60,000 samples and reports every locus with p < 5e-8 in any trait. LD pruning leaves 0.8 million effective independent SNP tests, and the trait correlation matrix leaves 5 effective independent trait tests. How many false-positive loci should the reported union be expected to contain?