How to Get a Bioinformatics Scientist Job

What the job actually is, and the four job families hiding behind one title

"Bioinformatics Scientist" is not one job. It is at least four, and the biggest single cause of wasted applications is pursuing the wrong one. Read the requisition's responsibilities section before the requirements section. The responsibilities tell you which family it belongs to, and the family tells you whether your background clears the screen.

The first family is the embedded analyst. You sit inside a research programme, a disease area, or a platform team, and you are the person who turns a sequencing run into a decision. The work is experimental design conversations, QC, differential analysis, figures that go into a go/no-go deck, and arguing with a bench scientist about whether a batch effect is real. Titles: Bioinformatics Scientist, Computational Biologist, Scientist I or II. This family is the most likely to require a PhD, because the value is scientific judgement rather than throughput.

The second family is the pipeline and platform builder. You own the workflows, the compute, the reference data, the containers, and the thing that lets twenty other people run an analysis without asking you. Titles: Bioinformatics Engineer, Computational Biology Software Engineer, Scientific Software Engineer, Genomics Platform Engineer. This family is much less credential-gated. Real engineering practice and a repository that proves it will usually beat a PhD with unrunnable code, and senior pay is competitive with the Scientist track.

The third family is the methods developer. You build new models, new statistics, new algorithms, and you publish. Titles: Senior Scientist, Principal Scientist, Machine Learning Scientist, Research Scientist. This is PhD-gated in practice, usually with a publication record in the specific method area, and it is the smallest of the four by headcount.

The fourth family is clinical bioinformatics, inside a CLIA-certified and CAP-accredited laboratory in the US, or an accredited genomics laboratory hub in the UK. The work is variant calling and interpretation support, pipeline validation, and the documentation that proves the pipeline does what you claim it does. It is more procedural than research bioinformatics, it is more stable through funding cycles, and it is the only part of the field with a genuine credentialed ladder. Candidates who assume industry research is the only destination never look at it, which is part of why it is easier to enter.

Do you need a PhD, and where credentials actually gate you

For a large share of the jobs, no. For the job titled "Scientist" at a mid-size or large biotech or pharma company, usually yes, and not because the work needs it. Job architecture at those companies defines levels partly by degree. A Scientist I requisition will often read "PhD in bioinformatics, computational biology, genetics or related field, or MS plus 4+ years of experience". That second clause is real and is routinely honoured. People talk themselves out of applying because they stop reading at the first half of the sentence.

Where the PhD genuinely matters is where you will be the scientific owner of a question nobody has answered: target discovery, novel method development, anything whose deliverable is a hypothesis rather than a number. The PhD is a proxy for having survived a multi-year project that did not work for the first three years. Hiring managers are testing for that, and they will accept other evidence of it, but you have to supply the evidence rather than expecting them to infer it from a job title.

If you have a wet-lab PhD and have been teaching yourself computation, you are strong for family one and weak for family two. Lead with the biology and prove the code runs. If you have a CS or physics degree and no biology, you are strong for family two and will struggle in family one until you can speak fluently about library prep, UMIs, strandedness, duplicate reads, and why a sample failed. Neither gap is fatal. Pretending the gap does not exist is.

A postdoc is no longer the default industry entry ticket in computational biology. Companies hire directly out of PhD programmes into Scientist I, and a postdoc taken purely as a credential, with no new skill or publication attached, can read as drift. Take one because a specific lab will teach you something you cannot get elsewhere.

The credential picture changes completely in clinical work, and this is the part most career advice gets wrong. In the US, nobody licenses a bioinformatician, but directing a clinical genomics laboratory generally requires board certification from the American Board of Medical Genetics and Genomics, obtained through an ACGME-accredited Laboratory Genetics and Genomics fellowship that you can only enter with a doctoral degree. A bioinformatician inside such a lab works under that director and needs no personal certification.

In the UK, "Clinical Scientist" is a title protected by law and registered by the Health and Care Professions Council. The main entry route is the NHS Scientist Training Programme, a three year salaried training post in Clinical Bioinformatics (genomics, physical sciences or health informatics specialisms) that leads to HCPC registration. There is also an equivalence route through the Academy for Healthcare Science for people who already hold the competences from elsewhere. Numbers of STP posts are small and recruitment runs once a year on a fixed national cycle, so if that is your route, find the opening date before you plan anything else. Check the current requirements with the National School of Healthcare Science and the HCPC rather than relying on an article, including this one.

How hiring actually works, stage by stage

Who screens you first depends on company size. At a startup under about fifty people, the hiring manager reads every application personally, often within days, and a good repository link can get you a call in under a week. At a large pharma company a recruiter or an automated screen runs first against the requisition text, and the hiring manager sees a shortlist. At an academic or hospital core facility, HR screens against a rigid published qualification list and a missing stated requirement is an automatic rejection however good you are, so apply only where you can tick every literal line.

Stage one, recruiter screen, 20 to 30 minutes. Logistics, work authorisation, salary expectations, a surface check that you have handled the data types in the requisition. Have a number ready and anchor it on the posted band rather than on your current pay. Several US states and cities restrict employers from asking salary history at all, so you are usually within your rights to decline that specific question and give a range instead.

Stage two, hiring manager call, 45 to 60 minutes. This is the real first filter. They will pick one project from your resume and go deep: what was the question, what did you do, what did you decide, what would you do differently. If you cannot defend one project in that kind of detail, nothing later in the loop rescues it.

Stage three, technical exercise. Two common shapes. A take-home: here is a public or subsampled dataset, do an analysis, send the code and a short write-up, usually with a stated budget of four to eight hours. Or a live session: screen-shared coding on a data-wrangling problem plus reasoning questions about a messy experimental design. Take-homes are more common here than in general software hiring because the job is literally to produce a defensible analysis document. Ask what the time budget is, and treat an open-ended exercise with no budget and no feedback as a signal about how the team works.

Stage four, the seminar or chalk talk. Standard for PhD-level Scientist roles, common in biotech, near-universal in academia. Thirty to forty-five minutes on your own work to an audience that includes people who do not compute. It is scored on whether a bench scientist followed you, not on whether the methods slide was impressive.

Stage five, the panel. Four to six one-on-ones including at least one wet-lab collaborator and often a program lead or a data engineer. The wet-lab interview is the one candidates underprepare and the one that most often sinks them. Those people are deciding whether working with you will be bearable, and whether you will tell them their experiment is underpowered before they spend the money rather than after.

Stage six, references and offer. Reference checking in this field is frequently informal and lateral: the hiring manager knows someone who knows you. Assume anything you have said publicly about a previous employer is discoverable.

Two variants worth knowing. Clinical diagnostic labs add a stage on regulatory and quality awareness, because an unvalidated change to a production pipeline is a reportable event rather than a hotfix. CROs and service providers add a stage on client communication, because you will be on calls with the customer's scientists and the deliverable is their confidence as much as their data.

• Startup: days to three weeks. The hiring manager reads everything, and a repository link is the strongest single asset.
• Mid-size biotech: four to eight weeks, structured loop, seminar for Scientist titles.
• Large pharma: eight to sixteen weeks, recruiter and automated screen first, keyword alignment with the requisition matters more here than anywhere else.
• Academic or hospital core facility: slow, HR screens against a literal qualification list, seminar required, pay is public and lower.
• Clinical diagnostic lab: adds a quality and regulatory conversation, and values documentation discipline over novelty.

The portfolio: what to build, in order

This is the part of the field where a public portfolio genuinely substitutes for pedigree, and most candidates build the wrong thing. A folder of notebooks with no README, no environment specification, no random seed and no stated question is worse than nothing, because it demonstrates exactly the habit the hiring manager is screening out. Build fewer things, finished.

Project one: a workflow that runs. Pick one real analysis (bulk RNA-seq differential expression, a small variant-calling workflow, an ATAC-seq peak pipeline) and implement it in Nextflow, Snakemake or WDL. Containerise every step with Docker or Apptainer. Pin versions. Include a test profile that runs on a subsampled dataset in under ten minutes on a laptop, and wire it to CI so a passing badge on the README proves it still runs today. That badge does more work than any bullet point on your resume. For a shortcut to good practice, read how nf-core structures its pipelines and follow it. Contributing one reviewed nf-core module is a checkable public credential, and the review itself teaches you the conventions.

Project two: an honest reanalysis. Take a public dataset from GEO, SRA, ENA, recount3, the GDC, GTEx, ENCODE or the Human Cell Atlas, and answer one specific question. The README opens with the question, the data (how many samples, what library chemistry, what depth, what genome build), and the answer in two sentences. Then the QC decisions and why you made them, then the analysis, then the limitations. If the result is null, say so. A clean null result with a correct power argument is a stronger hiring signal than a p-value fished out of a confounded design, and experienced reviewers see the difference immediately.

Project three: a benchmark. Compare two tools or two parameter choices on data where the truth is known: Genome in a Bottle benchmark sets for small variants, simulated reads where you control the ground truth, or a spike-in. Report precision, recall and F1 stratified by region difficulty, report the compute cost, and state a conclusion with its caveats. This proves you can evaluate rather than only execute, which is the single hardest thing to assess from a resume and the thing that separates a scientist from a tool operator.

Project four, optional and high value: a contribution to something people use. A merged pull request to samtools, bcftools, scanpy, Seurat, a Bioconductor package, an nf-core module. A Bioconductor package of your own that passed review is a particularly strong externally validated signal, because the review is rigorous and public and you did not control it.

What not to build: a dashboard nobody asked for, a general-purpose "bioinformatics toolkit" with three functions, a reproduced Kaggle tutorial, or a deep learning model trained on a dataset far too small to support it. These read as portfolio-filling, and reviewers have seen all of them many times.

• Every repository: README with the question first, environment pinned, one command to reproduce, a licence, and a short statement of what you would do with more data or more time.
• Use real file formats and get them right: FASTQ, BAM, CRAM, VCF, GTF, BED. Coordinate-system errors in a public repository get noticed.
• Name the genome build everywhere (GRCh38, T2T-CHM13, GRCm39) and the annotation release. Omitting it is a tell.
• If the data is controlled-access (dbGaP, UK Biobank, All of Us), do not put it in a public repository. Put the code there and say where the data lives and under what agreement.

The resume: what is read and what is skipped

A bioinformatics resume gets about forty seconds from a computational hiring manager on the first pass. They are looking for four things: the data types you have handled, the scale, what you decided, and whether you can ship. Everything else is noise to them, though some of it still has to be present for the keyword screen at larger companies.

Skipped almost entirely: a twenty-item skills block; "proficient in Python and R"; coursework; a generic objective statement; a publication list with no indication of what you did; tool logos; GPA after your first job.

Read closely: bullets shaped as "analysed X samples of Y data to answer Z, which led to decision W". Concrete scale (96 RNA-seq libraries, 40,000 cells, 1,200 exomes, 30x WGS on 300 samples). A named compute environment (Slurm, AWS Batch, AWS HealthOmics, GCP, Terra, DNAnexus, Seqera Platform). A named workflow language. And a GitHub link that is not empty.

For publications, state your contribution in a parenthetical. "Co-first author; built the single-cell pipeline and performed all differential abundance analysis" tells a hiring manager far more than the journal name. If you are sixth author on a consortium paper, say what you did or compress it into a list.

For the applicant tracking screen at larger companies, mirror the requisition's own nouns. If the posting says "single-cell RNA-seq", do not write only "scRNA-seq". Write both, once each. This is mechanical, it costs nothing, and it is the difference between a human seeing your application and not.

If you are transitioning from academia, cut the CV. Four pages with teaching, conferences and posters is correct for a faculty application and wrong for industry. Two pages maximum, projects reframed as deliverables, methods reframed as decisions, and the grant you were funded by removed unless the funder is the hiring company's partner.

What the interview really tests

Almost nobody is testing whether you can implement a sorting algorithm. The technical bar in bioinformatics interviews is reasoning under messy data, not algorithmic puzzles, and the failures cluster in four places.

Experimental design and confounding. "All the treated samples were run in batch two." What can you conclude, what can you not conclude, what would you ask for. This is the most common question and the most common failure. The right answer usually starts with "nothing cleanly, and here is why", not with a tool name.

Statistics as applied to this data. Multiple testing, and why FDR rather than Bonferroni for a 20,000-gene screen. What a p-value histogram should look like, and what a spike near 1 or a hump in the middle is telling you. Pseudoreplication in single-cell data: cells from one animal are not independent replicates, and the usual correct move is pseudobulk aggregation to the sample or donor level before differential testing. Normalisation choices and their assumptions. Power, calculated before the experiment rather than after.

File and reference literacy. 0-based half-open BED versus 1-based inclusive GTF, VCF and SAM. Strandedness and what happens when you set it wrong. Which transcript set you used and why MANE Select exists. Build mismatches and the hazards of liftover. Alt contigs and multi-mapping. This is not trivia. Wrong answers here corrupt results without raising an error, which is exactly why they are asked.

Debugging and ownership. "The pipeline ran green and the results look wrong. What do you check, in order?" They want a systematic answer: inputs and sample sheet first, then QC metrics, then a known-positive control, then one sample traced end to end by hand. A candidate whose first move is "I would rerun it" is being screened out.

Then the non-technical half, which decides more offers than the technical half. Can you explain a result to a bench scientist without condescension. Will you push back on a bad design before the money is spent. Can you be told your analysis is wrong without defending it. Hiring managers in this field describe a large part of the job as translation between the bench and the analysis, and they interview for it deliberately.

Prepare one project you can defend in complete depth, including what went wrong, what you got wrong, and what you changed. Candidates who present only successes read as either junior or evasive.

• Expect live coding at the level of: group and summarise a table, join two files on identifiers that do not quite match, parse a VCF field, write a function with a test. Pandas, dplyr, awk, occasionally SQL.
• Expect a take-home that is really a writing test. The write-up is scored as heavily as the code.
• Expect at least one question about reproducibility: could you rerun this analysis a year from now, and how.
• Expect one question about a time you disagreed with a wet-lab collaborator. Have a real one ready, with the outcome.

Where the jobs are in 2026 and 2027, and how to find them

The market is not what it was at the 2021 funding peak, and pretending otherwise helps nobody. Early-stage biotech hiring tightened after that peak and has not returned to it, and layoffs have been a regular feature of the sector since. What has held up better: clinical diagnostics and reference laboratories, where sequencing volume keeps growing; CROs and service providers, who absorbed work companies stopped doing in house; large pharma computational groups, slower to hire and also slower to cut; and the platform, instrument and tooling companies that sell into the sector regardless of who is doing the science.

That makes company selection part of the job search rather than an afterthought. Before you apply to a private company, check when it last raised and what it said the money was for, and check whether its lead programme has read out. Before you apply to a public one, the cash position and runway are in the filings. A role at a company with eight months of runway is a different decision from the same role at one with three years, and nobody will tell you which you are looking at unless you ask.

Geography still concentrates: Boston and Cambridge, the San Francisco Bay Area, San Diego, Seattle, Research Triangle, and the Maryland and DC corridor in the US; Cambridge, Oxford and London in the UK; Basel, Munich, Heidelberg, Copenhagen and Barcelona in Europe. Fully remote computational roles exist and are a real fraction of postings, but they skew senior, because the job depends on knowing what the bench is actually doing. Early in your career, physical proximity to a wet lab accelerates you, and hiring managers know it.

Where to look, concretely: company career pages directly (the aggregators lag), the Biostars and Bioconductor job boards, the nf-core and scverse community channels, Nature Careers and the society job boards (ISCB, ASHG), and the NHS Jobs site plus the national STP recruitment cycle if you are in the UK. Set alerts on company pages for the ten companies whose science you can actually discuss, and check them weekly. That beats volume applying by a wide margin.

Two underrated entry points. Core facilities at universities and hospitals hire continually, pay less, and give you breadth across assay types faster than any industry role will; two years there is a strong launchpad into industry. And instrument and reagent vendors hire field bioinformatics and application scientists, which puts you in front of dozens of labs a year and builds a network no internal role can match.

Work authorisation is a real variable in this field, because a large share of the candidate pool trained on student visas. Ask about sponsorship in the recruiter screen rather than at offer stage. Some employers sponsor routinely, some never do, and some will say so in the posting. Immigration rules and costs move, so check the current position with the employer and with official guidance rather than with an article or a forum thread. In the UK, note that NHS Scientist Training Programme posts and university posts have their own sponsorship rules that differ from private sector ones.

Areas worth orienting a portfolio towards, because the analysis is unsettled and the candidate pool is thinner than for bulk RNA-seq: single-cell and spatial transcriptomics (Visium, Xenium, MERFISH, CosMx); long-read sequencing (Oxford Nanopore, PacBio HiFi) for structural variants, methylation and hard regions; proteomics, especially DIA mass spectrometry, which postings ask for far more often than candidates list it; multi-omic integration; and clinical variant interpretation at scale. Immune repertoire analysis and cell therapy manufacturing analytics are both worth a look and rarely appear in career advice.

Getting the first job with no industry experience

Coming out of a PhD, a master's programme, or a career change, the order of operations that works is: finish one portfolio project properly, rewrite the resume around deliverables, then apply narrowly rather than broadly. Two hundred generic applications produce fewer interviews than twenty where you have read the company's papers and can name the data type they work with and the question it answers.

Referrals matter more here than in most fields because the community is small. Practical routes in: answer questions on Biostars or the Bioconductor support site under your real name; contribute to an open pipeline project and interact with the maintainers in public; present a poster at a meeting where industry recruits (ASHG, AGBT, ISMB, RECOMB, and the specialist meeting for your assay type); attend an nf-core hackathon, which are online and free and put you in a room with people who have open requisitions. None of this is networking theatre. It is doing visible work next to people who hire.

Internships and co-ops remain the highest-conversion route for students, and many large pharma and biotech companies run structured computational internships that convert to offers. Apply roughly nine to twelve months ahead. These close earlier than people expect.

Switching in from software engineering: take one serious genomics or molecular biology course and build the benchmark project. Your engineering practice is already ahead of most of the field. What you need to prove is that you will not produce a fast, clean, confidently wrong answer because you did not know what a duplicate read, a strandedness setting or a batch effect does to the result.

Switching in from the wet lab: your advantage is that you know what the data actually is and most of your competition does not. The gap is engineering credibility, and it closes with one public repository containing a workflow that runs reproducibly with tests. Target embedded analyst roles where biological judgement is the product, not platform roles where you will be compared against software engineers.

One more route people miss: the clinical laboratory. Genomics laboratories hire variant analysts and clinical bioinformatics staff continuously, train on the job, and value documentation discipline over novelty. It is one of the few places where a careful person with an MS and no industry history gets hired on their first application.

What a screen is looking for

These are the terms that a resume screen, human or automated, is matching against for this role. Use the ones that are true of you, in the words the posting uses.

• bioinformatics
• computational biology
• clinical bioinformatics
• NGS
• next-generation sequencing
• RNA-seq
• single-cell RNA-seq
• scRNA-seq
• whole genome sequencing
• WGS
• whole exome sequencing
• WES
• variant calling
• variant interpretation
• GATK
• samtools
• bcftools
• BWA
• STAR
• salmon
• DESeq2
• edgeR
• limma
• Seurat
• scanpy
• Bioconductor
• pseudobulk
• Nextflow
• nf-core
• Snakemake
• WDL
• Cromwell
• Galaxy
• Docker
• Singularity
• Apptainer
• Python
• R
• Bash
• SQL
• Git
• GitHub Actions
• CI/CD
• Slurm
• HPC
• AWS
• AWS Batch
• AWS HealthOmics
• Google Cloud
• Terra
• DNAnexus
• Seqera Platform
• ATAC-seq
• ChIP-seq
• methylation
• long-read sequencing
• Oxford Nanopore
• PacBio HiFi
• spatial transcriptomics
• Visium
• Xenium
• proteomics
• DIA mass spectrometry
• multi-omics
• GRCh38
• T2T-CHM13
• MANE Select
• VCF
• BAM
• CRAM
• GTF
• BED
• FASTQ
• statistical genetics
• GWAS
• differential expression
• batch effect
• FDR
• multiple testing correction
• benchmarking
• reproducible research
• pipeline validation
• machine learning
• AlphaFold
• protein language model
• ESM
• AlphaMissense
• SpliceAI
• CADD
• ACMG/AMP
• ClinGen
• CLIA
• CAP
• ISO 15189
• HCPC
• Scientist Training Programme
• GEO
• SRA
• TCGA
• GTEx
• dbGaP
• UK Biobank
• PhD computational biology
• MS bioinformatics

Mistakes that cost people this job

Questions people ask