Appendix F — Cross-topic problems and case studies

NoteCredit

The problems and case studies in this appendix are adapted from Matteo Tadiello, Modern Bioinformatics: From Foundations to Frontiers (CC BY 4.0), used with permission.

The review problems at the end of each chapter test one topic at a time. In the oral exam, questions often move from one topic to another: from a sequence search to evolution, from structure to function, from an experiment to its statistics. The problems in this appendix combine several chapters and have no single correct answer. Use them to practise explaining a complete line of reasoning: what the biological question is, which data and methods would address it, what each result can and cannot show, and how the conclusion would be validated.

F.1 Cross-topic problems

These problems integrate concepts from at least three chapters. For each problem, define the biological claim, identify the data-generating process, justify the computational formulation, state assumptions, compare alternatives, and propose validation.

  1. You discover a previously unannotated bacterial gene associated with antibiotic resistance. Design the sequence, structure, evolutionary, expression, and experimental workflow needed before making a functional claim.
  2. A profile HMM and a structure search disagree about a remote homolog. Explain possible causes and how you would adjudicate them.
  3. A gene is differentially expressed but its protein abundance is unchanged. Give technical and biological explanations and design follow-up tests.
  4. A pangenome graph improves variant recall. Explain how you would demonstrate that the gain is real rather than benchmark leakage or truth-set bias.
  5. A high-confidence predicted protein complex has no AP-MS support. Reconcile the evidence and propose orthogonal validation.
  6. A gene family tree conflicts with the accepted species tree. Compare duplication/loss, incomplete lineage sorting, horizontal transfer, recombination, and artefact explanations.
  7. A classifier predicts disease from transcriptomes with AUROC 0.97 but fails prospectively. Diagnose data leakage, batch structure, spectrum effects, calibration, and target shift.
  8. An RNA-seq study has three treatment samples processed on one day and three controls on another. Explain what can and cannot be recovered computationally.
  9. A protein has 35% sequence identity to a characterised enzyme. Build an evidence hierarchy for transferring catalytic function and substrate specificity.
  10. Compare the meaning of an E-value, adjusted p-value, posterior probability, bootstrap proportion, predicted confidence score, and calibrated class probability.
  11. Design a database record that distinguishes raw evidence, inferred annotation, version, uncertainty, and provenance.
  12. Explain how repeats affect read mapping, genome assembly, gene annotation, expression quantification, and evolutionary inference.
  13. A rare cell population appears only after single-cell integration. Explain how to decide whether it is biology or an integration artefact.
  14. You need to identify direct interaction partners of a membrane protein in living cells. Compare experimental and computational options.
  15. A foundation model produces accurate embeddings. Explain what benchmarks would establish biological utility without circularity.
  16. Design a reproducible multi-omics study from sampling through public deposition, including negative controls and validation.

F.2 Case study A: from sequence to mechanism

This case study traces a biological question from comparative genomics through molecular mechanism. The routes can be analysed sequentially or assigned as separate problem sets.

F.2.1 The problem

A conserved protein of unknown function is found in several pathogenic bacteria but is absent from closely related non-pathogens. Design a strategy to determine its origin, molecular function, structure, interaction partners, and contribution to pathogenicity.

Route 2: evolution and phylogeny

  • How would you distinguish vertical inheritance, loss, and horizontal transfer?
  • What alignment and substitution-model problems could distort the tree?
  • How do you reconcile a gene tree with a species tree?
  • What evidence would support selection without overinterpreting dN/dS?

Route 3: structure and function

  • What does a high-confidence predicted fold establish?
  • How would you evaluate a putative active site or ligand-binding pocket?
  • When is docking informative, and how would you validate it?
  • How would disorder, oligomerisation, or membrane localisation change the plan?

Route 4: expression and interactions

  • Design a condition-aware expression experiment with biological replication.
  • Choose an interaction assay and explain its false-positive/negative profile.
  • What would count as orthogonal evidence for a direct interaction?

Route 5: integration

  • Build the strongest defensible conclusion from mixed evidence.
  • Identify which claims remain predictive rather than causal.
  • Define the minimum experiments needed before proposing a drug target.

F.2.2 Evaluation criteria

A rigorous analysis connects each biological claim to the evidence generated by the corresponding method, states algorithmic assumptions, distinguishes estimated confidence from correctness, and proposes independent validation. A tool list without an explicit inferential argument is insufficient.

F.3 Case study B: from cohort to biological claim

This case study develops an end-to-end analysis of a heterogeneous clinical cohort, from experimental design and data generation to prediction, causal interpretation, and prospective validation.

F.3.1 The problem

A cohort study collects genomes, bulk RNA-seq, single-cell RNA-seq, and plasma proteomics from patients and controls across three hospitals. The goal is to discover a disease mechanism and a predictive biomarker. Design the analysis and justify each inferential step.

Route 1: design and data generation

  • What replication and randomisation are required?
  • Which variables must be balanced or recorded?
  • How do sequencing and mass-spectrometry error processes differ?
  • What quality controls detect sample swaps and contamination?

Route 2: genomic analysis

  • Choose mapping/reference representations and variant-calling strategies.
  • Explain reference bias, population structure, phasing, and structural variants.
  • How would you validate a candidate regulatory variant?

Route 3: expression and proteomics

  • Specify the bulk RNA-seq design matrix and contrasts.
  • Explain single-cell pseudoreplication and a defensible differential-expression unit.
  • Describe peptide-spectrum matching, FDR, protein inference, and missing values.
  • How would you interpret RNA–protein disagreement?

Route 4: machine learning

  • Define the unit of splitting and prevent patient/site leakage.
  • Choose metrics for a rare disease and assess calibration.
  • Compare L1, L2, tree-based, and representation-learning approaches.
  • How would you perform nested model selection and external validation?

Route 5: integration and causality

  • Compare early, intermediate, and late multi-omics integration.
  • Separate a predictive biomarker from a causal mechanism.
  • Explain how site, ancestry, treatment, and cell composition could confound the result.
  • Specify an orthogonal experiment and a prospective validation study.

F.3.2 Evaluation criteria

The answer must preserve the experimental unit, connect data-generating processes to statistical models, prevent leakage, control multiplicity, distinguish prediction from causation, and specify independent validation.