--- module: 102-02 language: en chapter: 102 title: "Experimental Biomedical Methods, Measurement, Omics, and Reproducibility" module_title: "Molecular, genetic, omics, single-cell, and computational methods" source_sha256: 49f97f53a8492d636836fe79b64a49f3dc58260bbaa77e2623888fa6a6257528 --- # Molecular, genetic, omics, and computational methods ## Specimen versus representation ### Extraction, amplification, labelling, filtering transform #### Each step can lose or enrich material #### Controls must follow the whole chain ### Extraction: lyse, inactivate nucleases, recover #### Yield, purity ratio, integrity differ #### RNA changes during ischaemia and handling #### Blanks catch contamination; spike-ins track recovery #### Kit differences can resemble biology ## PCR and quantification ### Denaturation, primer annealing, extension ### End-point PCR poorly quantitative after saturation ### Quantitative PCR uses threshold-crossing cycle #### Valid only with sound efficiency and baseline ### Reverse transcription varies with enzyme and priming ### Reference genes must be stable in the experiment ### Digital PCR counts positive partitions #### Probability model estimates molecule number ## Sequencing ### Sequencing by synthesis on clonal templates ### Short reads accurate but ambiguous in repeats ### Long reads span structural variants and isoforms ### Coverage improves low-frequency variant detection #### Cannot repair systematic mapping or amplification bias ### Workflow from library preparation to annotation #### Index misassignment moves reads between samples #### Duplicate reads may reflect amplification #### Reference bias misses novel sequence #### Orthogonal confirmation when consequences are high ## Detecting variants and associations ### Sanger for small regions and confirmation #### Consensus signal misses low-frequency variants ### Chromosome analysis: large abnormalities, dividing cells ### Fluorescence in situ hybridisation #### Selected loci in cells or tissue ### Copy-number methods may miss balanced rearrangements ### Method follows variant type and question ### Genome-wide association tests common variants #### Linkage disequilibrium: marker may only tag #### Control structure, relatedness, multiple testing #### Fine mapping narrows; experiments test mechanism ### Polygenic scores weaken in underrepresented groups ## Genome editing ### Guide directs nuclease cutting #### Repair gives indels or incorporates template ### Base and prime editors avoid double-strand break ### Measure off-target, mosaicism, rearrangement, p53 ### Rescue or independent guides confirm the target ## Transcripts, proteins, metabolites ### RNA-seq counts depend on length, depth, mapping ### Bulk change: within-cell or cell-mixture shift ### Transcript does not equal protein or activity #### Translation, degradation, modification intervene ### Proteomics digests to peptides for tandem spectra #### Isoforms need distinguishing peptides ### Metabolites shift fast with diet, time, handling #### Many untargeted features stay unidentified #### Identify by mass, retention, fragments, standards ## Epigenomic and single-cell views ### Bisulfite conversion distinguishes modified cytosine ### Chromatin immunoprecipitation needs specific antibody ### Accessibility enzymes enter open chromatin ### Single-cell barcodes sample transcriptomes sparsely #### Capture, dissociation, viability bias cells seen #### Ambient RNA, doublets, dropout #### Clusters are analytic and need validation ## Spatial and multi-omic integration ### Spatial methods preserve location #### Resolution, sensitivity, segmentation trade off #### Deconvolution depends on references and models #### Proximity is not proof of communication ### Multi-omics multiplies dimensions and missingness #### Correlated features let many models fit ### Pathway enrichment generates hypotheses #### Selecting and testing on same data is circular ## Batch effects and pipelines ### Date, lot, instrument, operator differ systematically ### Batch confounded with group is uncorrectable ### Randomise across batches with bridge controls ### Correction can remove biology or fake similarity ### Pipelines are part of the instrument #### Versions, references, seeds change output ### Prespecify QC; rebuild from raw inputs ## From data to causal inference ### Bottleneck shifts to causal interpretation ### Signatures can classify without mechanism ### Enrichment may reflect composition or stress ### Discover, validate, localise, perturb, link phenotype ### Omics does not repeal controls and replication