Bioinformatics & omics

Thirteen subjects about measurement, and they agree on one thing: each omics layer is shaped by the physics of its own measurement, and the artefacts follow from it. Proteins cannot be amplified, which is why proteomics is harder than genomics. Metabolites share no common backbone, so no single method sees the layer at all. Every artefact of single-cell data — dropout, doublets, ambient RNA — is a consequence of how the cell got into the tube. In cryo-EM the microscope is rarely the limit; the ice and the molecule’s own flexibility are.

The second idea: seeing where something is costs you what it is, and the other way round. Spatial capture is unbiased but mixes cells; imaging is precise but sees only a panel.

The third: inference has a boundary worth stating. Stacking omics layers multiplies noise more often than signal; a model trained on sequence alone learns structure until it stops; DNA is an extraordinary archive and a poor computer.

Start with bioinformatics and multi-omics: it asks why combined layers so rarely multiply signal, and that question frames every method page here.

Bioinformatics and multi-omics The statistical structure of omics integration — different noise models per layer, missingness that is not random, batch effects confounded with biology, and the sample sizes the arithmetic actually demands.
Biocomputing and DNA computing Why DNA storage wins on density and loses on write cost and access latency — synthesis and sequencing error models, random access by PCR, and an honest comparison against magnetic tape.
Biological AI and bioinformation services The evolutionary signal that makes protein language models and structure predictors work — coevolution in multiple sequence alignments — and the failure modes: leakage, orphan families, conformational states, and the gap between predicting and designing.
Genomics and DNA design Designing sequence to be built and to work — the synthesis constraints of repeats, GC extremes and secondary structure, why codon optimisation is a real but overstated lever, and how regulatory elements set expression.
Long-read sequencing Nanopore ionic current and PacBio circular consensus as distinct error models, why read length rather than accuracy resolves repeats, structural variants and phasing, and what raw versus consensus accuracy means.
Metabolomics Chemical diversity as the defining constraint of metabolite measurement: extraction and platform coverage, annotation as the real bottleneck and what confidence levels mean, matrix effects, and why quenching at sampling is mandatory.
Mass-spectrometry proteomics The dynamic range of plasma, stochastic under-sampling in data-dependent acquisition versus DIA, and why peptide measurements do not cleanly reconstruct protein-level truth.
Single-cell omics Droplet capture and barcoding, why a zero is usually sampling depth rather than biological absence, doublets and ambient contamination, and the stress signature that dissociation itself induces.
Spatial epigenomics In-situ transposition and antibody-tethered cleavage, deterministic microfluidic barcoding, why accessibility and histone-mark data are extremely sparse per pixel, and what this measures that expression does not.
Spatial transcriptomics and spatial multi-omics The resolution-versus-capture trade in spatial gene expression, spot-level mixtures and deconvolution, targeted imaging methods and their panel limit, and how protein layers are added.
Structural biology and cryo-electron microscopy Vitrification and why crystalline ice ruins an image, the air-water interface as the dominant practical problem, preferred orientation, heterogeneity as the real resolution limit, and what a resolution figure does and does not mean.
Toxicity screening and toxicogenomics Transcriptomic signatures as a hazard proxy, benchmark dose from gene-set responses, why in-vitro concentration is not in-vivo dose, and what the Tox21 and new-approach-methodology programmes established.