The falling cost of DNA sequencing is one of the most cited technology curves of the last twenty years, and it is genuinely remarkable.
What it obscures is that the expensive part moved. Reading a genome is now a routine laboratory operation. Knowing what the reading means is a skilled human judgement that has not become cheaper at anything like the same rate.
The cost curve that moved and the one that did not
Sequencing is now fast enough and cheap enough that organisations generate far more data than they can interpret. That is true in hospitals, in agriculture, in industrial strain development and in research.
The backlog is not computational. Alignment and variant calling are largely solved and automatable, and that layer sits in the bioinformatics direction. The backlog is the step after: given a difference between this sequence and a reference, what does it actually mean, and what should anyone do about it.
Every organisation that bought a sequencer discovered the same thing. The instrument was the affordable part.
That question needs biology, statistics, familiarity with the relevant databases and literature, and in clinical settings a formal framework with regulatory weight behind it. It is not a button.
What the direction covers
The scope: inheritance, mutations, sequencing, the human genome, population genetics and genomic medicine.
In practice that resolves into four capabilities.
Understanding what sequencing produces. What the platforms do, what their error profiles are, what coverage means, and which questions a given method can and cannot answer. Someone who does not know the failure modes of the technology will trust a result they should not.
Variant interpretation. The core skill. Classifying a difference as benign, pathogenic or uncertain, and being honest about how often the answer is genuinely uncertain.
Population genetics. Allele frequencies, structure, linkage. The statistical basis for almost every claim made about groups rather than individuals, and the layer where most public overstatement happens.
Applied genomics. Genomic medicine, agricultural breeding, industrial strain characterisation. The same foundation pointed at different problems.
Why interpretation is the bottleneck
Three reasons, and they compound.
Most variants are of uncertain significance. A sequencing run returns a large number of differences and only a minority have well-established meaning. The honest output is frequently "we do not know", and producing a defensible "we do not know" takes as much work as producing an answer.
The evidence base moves. A variant classified one way three years ago may be reclassified as evidence accumulates. That means interpretation is not a one-time task and people doing it have to stay current, which is a continuous learning problem rather than a course.
Context determines meaning. The same variant means different things depending on the clinical picture, the population background, the other variants present and what is being asked. Interpretation without context is pattern matching.
The result is a role that cannot be automated away as easily as the field expected, and that has a genuine supply constraint.
Where this sits in the domain
Genetics and genomics is the third of ten directions in Astra Trainer's biotechnology domain. It sits directly alongside bioinformatics and computational biology, which covers reading genomes, sequences and proteins at scale, and the two are frequently scoped together because organisations need both halves and usually have one.
It also feeds the precision medicine and advanced diagnostics direction in the medicine and healthtech domain, for partners whose genomics work is clinical rather than industrial. Programs are sequenced with the partner's own scientists. You can see the ten directions here.
The roles, named
Clinical scientists in genomics. Interpreting and reporting variants in a diagnostic setting. Credentialled, regulated, and short almost everywhere.
Variant curators and scientific curators. Building and maintaining the evidence base that interpretation draws on. A growing role and a good entry point.
Genomic laboratory scientists. Running sequencing operations, managing quality, troubleshooting library preparation and coverage problems.
Genetic counsellors. Communicating results to patients and families. A regulated profession with its own qualification route, and in short supply in many health systems.
Plant and animal breeding scientists. Marker-assisted and genomic selection in agriculture, which is a large and less-discussed employer.
Strain characterisation scientists. In industrial biotechnology, understanding what a production organism's genome says about its behaviour.
Structural context: the US Bureau of Labor Statistics projects employment of medical scientists growing about 9 percent between 2024 and 2034, above the average occupation, and genomics is one of the areas pulling that.
Who can be trained into it
Two main pools, needing opposite halves.
Laboratory staff already running sequencing. They know the technology, the sample handling, the quality problems and the failure modes. What they lack is the interpretation framework and the statistical grounding. This is the shorter conversion and the more overlooked one, because these people are usually classified as technical rather than scientific staff.
Analysts from data backgrounds. Statistics, data science, occasionally software. They handle the data comfortably and lack the biology, which means they can produce an answer without being able to sanity-check it. This conversion takes longer than people expect and is worth doing properly.
Two smaller pools worth knowing about.
Clinical laboratory scientists from other disciplines, who already work inside accreditation and quality frameworks.
Agricultural and plant science staff, for whom genomic selection is an extension of breeding work they already understand.
The credential boundary is sharp here. Clinical variant interpretation and genetic counselling are regulated professional activities with defined qualification routes, and diagnostic laboratories operate under accreditation with documented competency requirements. Training builds the scientific capability and prepares people for those routes. It does not qualify anyone to issue a clinical report or to counsel a patient, and in this field the consequence of blurring that line falls on a patient.
Where the hard limits are
Two things a workforce program in this area should teach explicitly rather than leave to be discovered.
Uncertainty is the normal result. A person trained to always produce a classification will produce wrong ones. The professional skill includes returning "uncertain significance" with confidence and explaining what would change it.
Population claims are easy to overstate. Genomics has a long history of results about groups being reported with more confidence and more meaning than the underlying statistics support, in both scientific and public contexts. Population structure, allele frequency differences and ancestry inference all have real technical content and real limits. Anyone working with population-level data should understand what those methods do and do not establish, and that belongs in the training rather than in a later correction.
What to take from this
The cheap part is generating sequence. The constrained part is deciding what it means, and that gap is widening as sequencing volume grows.
Interpretation resists automation because most variants are uncertain, the evidence base moves, and meaning depends on context.
The two training pools need opposite halves, and the laboratory staff already running your sequencers are the shorter conversion and the one most often missed.
Clinical interpretation and genetic counselling are credentialled activities and training prepares people for those routes rather than replacing them.
And teaching the limits, particularly around uncertainty and population-level claims, is part of competence rather than a caveat attached to it.
If sequencing is cheap, where is the shortage?
In interpretation. Generating data is routine; deciding what a variant means requires biology, statistics, current literature and context, and it has not become cheaper at the same rate.
Can variant interpretation be automated?
Partly. Alignment and variant calling are largely automated. Classification resists it because most variants are of uncertain significance, the evidence base changes, and meaning depends on clinical and biological context.
Who can be trained into genomics roles?
Laboratory staff already running sequencing, who know the technology and need the interpretation framework, and analysts from data backgrounds, who handle the data and need the biology. The first is the shorter conversion.
Does training qualify someone to report clinical results?
No. Clinical variant interpretation and genetic counselling are regulated professional activities with their own qualification routes, and diagnostic laboratories carry accreditation and competency requirements.
Where does this fit in the domain?
Third of ten directions in Astra Trainer's biotechnology domain, usually scoped alongside bioinformatics. You can see them here.