Processing

• Normalisation choices — TMM, quantile, VSN — and their differing assumptions
• ComBat and surrogate variable analysis for batch correction, and over-correction risk
• Filtering low-expression features before, not after, statistical testing

Integration

• Concatenation versus multi-omics factor models such as MOFA
• Sample matching, scale mismatch and the dominance of the largest layer
• Interpreting a latent factor without inventing a biological story for it

Modelling

• Regularised models and tree ensembles when p greatly exceeds n
• Nested cross-validation — feature selection inside the fold, never outside
• Multiple testing control with Benjamini-Hochberg and reporting effect sizes

Diagnostics

• Reducing a signature to a panel that a PCR or targeted assay can measure
• Primer design constraints when the analytical target comes from omics data
• Analytical validation: sensitivity, specificity and independent cohort confirmation