An essay prompted by Alan Kay's 2011 talk on why the Internet and living systems scale while most software does not, with his numbers checked against real sources, and what the argument implies for a personal molecular data system built from a genome, a blood panel, and a glucose trace.
A working guide to diagnosing and handling batch effects in bulk RNA-seq, from design and TMM normalization through limma, ComBat-seq, and RUV, with the checks that tell you whether the correction helped.
What biological age tests measure, which clocks are worth running, how to compute them yourself from IDATs and a blood panel, and how much of the number is noise.
An essay prompted by Carolyn Bertozzi's lecture on bioorthogonal chemistry: what the reactions are, the second-order rate constants that decide what can be imaged in a living animal, and why the glycome is still a missing data layer in most molecular profiles.
How to use R and Bioconductor as the annotation, statistics, and integration layer for a personal molecular dataset: VCFs, RNA-seq counts, proteomics, and continuous glucose data, with the parts you should not do in R.
Why the data structure chosen for a genome, a variant set, an expression matrix, or a phylogeny decides which questions can be asked and what runs in reasonable time, prompted by a Boundary talk on coding agents.
A technical account of the privacy properties of consumer and clinical DNA testing: what data exists, who holds it, what re-identification attacks work, and a concrete setup for keeping your own sequence data under your control.
In the US, FreeStyle Libre 2 and 3 still require a prescription; Libre Rio, Dexcom Stelo, and Abbott Lingo do not. What each one costs, how they differ, and how to get the raw glucose data out.
Genetic Genie's methylation panel reads about two dozen genotypes out of a consumer array file. This guide shows how to extract and verify those calls yourself, what they can and cannot tell you, and how to measure actual DNA methylation if that is what you were after.
Pharmacogenomic panels marketed for ADHD medication selection rest on thin evidence for stimulants. Here is what the genotypes predict, which variants have real pharmacokinetic support, and how to extract them from your own sequencing data.
A postprandial reading of 97 mg/dL is a normal, unremarkable value in a person without diabetes. Here is what the number tells you, what it does not, and how to read postprandial glucose properly from CGM or fingerstick data.
How to run a genome-wide association study in R end to end, from VCF to QC to association testing to polygenic scores, and what you can and cannot do with a single genome.
HIPAA compliance is a property of the entity holding your data, not of a sequencing assay. Here is what HIPAA does and does not cover for genetic data, what to read in a lab's contract instead, and how to take custody of your own FASTQ, BAM, and VCF files.
Dehydration concentrates plasma by a few percent, which is far too small to explain most ALT and AST elevations. Here is what usually causes them, how to draw a clean measurement, and how to read enzymes against your own genomic and proteomic baseline.
A working pipeline for taking Olink, SomaScan, or DIA-NN output into R: QC, missingness, normalization, and longitudinal within-person modeling of a single individual's plasma proteome.
A working pipeline from FASTQ files to gene-level counts, differential expression, and enrichment, using salmon, pytximport, and PyDESeq2, with the failure modes specific to blood RNA from a single person.
A post hoc |log2FC| > 1 filter is the wrong tool. Test against a fold-change threshold directly with DESeq2's lfcThreshold or edgeR's glmTreat, shrink effect sizes before ranking, and pick the threshold from your own replicate structure.
How Rob Phillips turns molecular biology into estimates and equations: order-of-magnitude reasoning, counting molecules with coin flips, the thermodynamic model of gene regulation, and what the same method would look like applied to a person's own data.
A practical pipeline for turning a tellmeGen raw data export into a build-verified VCF, running quality control, annotating it, and understanding what a genotyping array can and cannot tell you.
Arthur Mattuck's derivation of the Laplace transform as the continuous analogue of a power series, and where the transform does real work in pharmacokinetics, gene circuits, signaling, stochastic gene expression, and survival analysis.