The science of microbial forensics

Battlefield bioforensics is the application; the science underneath is microbial population genetics. The defence framing supplies the questions — where did this organism come from, is this sample connected to that one — but the answers are read out of evolutionary history written in genomes, the same history epidemiologists read in foodborne outbreaks and hospitals read in ward transmissions. The limits of the method are therefore not administrative but biological, and they are worth knowing before any report built on it is believed too much.

What a genome comparison establishes#

A bacterial genome accumulates mutations as it replicates, and for many pathogens most single-nucleotide differences between close isolates arise this way. The count of such differences between two samples works as a rough molecular clock: near-identical genomes point to a recent common ancestor, and the number of differences speaks to how recent. A phylogeny built from these differences places the case sample on a tree of reference isolates. What the comparison establishes is relatedness — that the sample belongs, or does not belong, to the lineage that also produced these other isolates. That is the entire evidentiary payload: attribution to a lineage, not to a laboratory, a country or an intent.

Why confidence is statistical and bounded#

Several quantities cap the resolution. Mutation rates are estimated from observation — long-term culture lineages, transmission chains — and they vary by species, by strain, even by growth conditions, so the clock carries error bars. Within a single infection the population is not a clone: a sample captures only part of its diversity, and stochastic bottlenecks at each transmission add differences that ancestry alone does not explain. Evolution is not strictly tree-like either — recombination and homoplasy, the same nucleotide change arising independently twice, put noise on top of inherited signal. The hardest limit is database coverage: a phylogeny can only place a strain relative to genomes that have been sequenced. An isolate just as close to the case sample but never collected would be invisible, so every attribution statement is conditional on what the reference set happens to contain. Formally the result is a likelihood ratio — how much more probable the observed similarity is under one origin hypothesis than under another — never an identification.

Provenance carries half the weight#

A genomic match is only as good as the physical evidence behind it. The forensic weight comes from a documented chain of custody — who collected the swab, into what container, under what conditions — because a sequence cannot testify about its own handling. Degradation further bounds what the molecule can say: nucleic acids fragment with time, heat, moisture and nuclease exposure, so field-collected material may yield partial genomes, and partial data widen every confidence interval. Ambient-temperature preservation chemistry exists precisely to slow this decay between collection and the laboratory.

The honest boundary#

None of this reads intent. Population genetics can say that two isolates are close relatives; it cannot, by itself, distinguish a deliberate release from a natural emergence, because nature also produces transmission chains of near-identical genomes. Where the technology article lists agencies, field kits and programmes, this page stands behind them: attribution confidence is a probability conditioned on reference data and on evidence handling, and any account that reports it as certainty has stepped outside the science.