Protein Design Scientist Job Description

A Protein Design Scientist builds proteins that do not exist in nature, or reshapes natural ones, so that a molecule binds a chosen target, speeds a reaction, assembles into a particle or holds a shape a program needs. The job sits between a GPU cluster and a lab bench. The designer states the problem in physical terms, runs generative models over it, picks a short list of candidates, and then shows through experiments that the molecules fold and behave the way the model said they would. The deliverable is a set of amino acid sequences with data attached, not a hypothesis or a screening hit list.

The design loop

The University of Washington's Institute for Protein Design describes its deep learning toolkit in three stages: backbone design, sequence design and structure prediction. A designer sketches or generates a backbone, the three-dimensional shape the finished protein should take. Sequence design tools then fill in the chemistry. The institute's ProteinMPNN, in its own description, "takes a protein structure as input and quickly identifies combinations of amino acid that are likely to fold into that shape." Prediction closes the loop: the institute's RoseTTAFold "uses deep learning to quickly and accurately model protein structures based on amino acid sequences alone," and the institute says it uses the tool to evaluate designs "so that only the most promising molecules are produced and studied in the lab."

That filtering step is a large part of the craft. Models will happily emit thousands of sequences. The scientist decides which handful is worth real money, writing scoring scripts, setting cutoffs on predicted confidence and interface quality, and throwing out designs whose predicted structure drifts from the intended backbone.

Then it has to be made

Designs become DNA. The scientist orders synthetic genes, expresses the constructs in bacteria or mammalian cells, purifies the protein, and measures what came out: binding affinity on a biolayer interferometry or surface plasmon resonance instrument, melting temperature, oligomeric state by size exclusion chromatography, activity in a functional assay. When a design matters, the team solves a structure by cryo-electron microscopy or crystallography to confirm the fold atom by atom. Until that data exists, a design is a prediction, and the prediction is not the product. Expect part of every week to be spent on the unglamorous mechanics of getting protein made: construct design, plasmid maps, expression troubleshooting, buffer screens and repeated purifications of the one candidate that keeps aggregating.

Who else is in the room

Protein designers rarely work alone. A typical week involves molecular biologists who clone and express the constructs, assay scientists who run the readouts, machine learning engineers who maintain the pipelines, and a program lead who cares about one therapeutic or industrial target. In a biotech company, successful designs move on to developability and process teams who ask whether the molecule can be manufactured. In an academic lab, the same design becomes a figure in a paper and, often, an open-source release.

People reach this work through structural biology, biochemistry, biophysics, chemical engineering or computer science, then pick up the other half on the job. The Bureau of Labor Statistics says biochemists and biophysicists "need a Ph.D. to work in independent research and development," that "many Ph.D. holders begin their careers in temporary postdoctoral research positions," and that "Bachelor's and master's degree holders qualify for some entry-level positions in biochemistry and biophysics." That maps well onto protein design teams, where a doctorate is the usual route into an independent design role and bachelor's or master's scientists often run expression, purification and assay work alongside the designers.

A postdoc is a common bridge into the role, and in this field it doubles as the portfolio. Published designs, solved structures and a software contribution carry weight because the work is visible: the tools are open, the benchmarks are public, and anyone can see whether a design was validated or only predicted.

The technical floor

Expect to write real code. Python is the working language, model work runs in PyTorch, and jobs are scheduled on GPU clusters or cloud instances, so comfort with the shell, Slurm or a cloud batch service and version control is assumed. A working designer needs the current design and prediction tools, including diffusion models for backbone generation, message passing networks for sequence design, and AlphaFold-family or RoseTTAFold structure prediction, well enough to know where each one fails. Classical modeling still shows up: Rosetta energy functions, docking, molecular dynamics and simple physics checks catch problems a learned model will miss.

The wet-lab half is equally concrete. Cloning, protein expression in E. coli and mammalian cells, chromatography, binding kinetics and basic structural methods are the vocabulary of every design review. Scientists who cannot read their own SEC trace end up dependent on someone else's interpretation of whether a design worked.

Credentials and habits

No license or certification is required for this work. What stands in for one is evidence: a thesis, preprints, a GitHub history, a design that someone else reproduced. Two professional habits matter beyond the science. The first is record keeping, because a design campaign that cannot be reconstructed cannot be defended in a patent filing or a paper revision. The second is biosafety and biosecurity literacy: institutional review, knowing which experiments need approval, and ordering synthetic DNA through providers that screen orders. In 2024, as a witness told a House Energy and Commerce subcommittee in December 2025, more than 150 signatories of the Responsible AI x Biodesign statement committed "to only purchase synthesized genes from providers that screen their orders and customers."

The Bureau of Labor Statistics does not publish a protein design occupation. The closest fit is biochemists and biophysicists, SOC 19-1021, an occupation the agency describes as using "computer modeling software to determine the three-dimensional structures of proteins and other molecules." In May 2025 that occupation held 33,830 jobs with a median wage of $127,410. The Occupational Outlook Handbook projects employment of biochemists and biophysicists to grow 12 percent from 2025 to 2035, which the agency calls much faster than the average for all occupations, with about 2,900 openings a year on average over the decade. In the scientific research and development services industry, BLS recorded 21,900 jobs in that occupation in May 2025.

Why the title appeared

Recognition for the field came in 2024. The 2024 Nobel Prize in Chemistry "was awarded with one half to David Baker 'for computational protein design' and the other half jointly to Demis Hassabis and John M. Jumper 'for protein structure prediction,'" the American Institute of Physics noted on Oct. 9, 2024. Baker, as that announcement states, is "a professor of biochemistry, HHMI investigator, and the director of the Institute for Protein Design at the University of Washington School of Medicine."

That institute, established in 2012, now "brings together over 250 members with expertise in artificial intelligence, biochemistry, genome sciences, biological structure, pharmacology, immunology, computer science, and medicine." The institute says it played a key role in developing SKYCovione, which it calls "the first medicine ever created using computational protein design," a COVID-19 vaccine "featuring 60 copies of the SARS-CoV-2 Spike protein receptor binding domain" that received an emergency use listing from the World Health Organization. Employers for the work include pharmaceutical and biotechnology companies, startups built around a design platform, national labs and university institutes.

The policy backdrop to watch

Because designed sequences have to be synthesized to exist, screening of synthetic DNA orders is a regulatory pressure point. Executive Order 14292, issued May 5, 2025, directed the Office of Science and Technology Policy to revise or replace the 2024 Framework for Nucleic Acid Synthesis Screening. According to the Congressional Research Service, as of its August 3, 2026 update, the policy the White House released in July 2026 "replaced the 2024 policy; it did not revise or replace the 2024 Framework for Nucleic Acid Synthesis Screening." In June 2026 a group of researchers, AI companies and biotechnology developers called on legislators to make screening of orders for synthetic nucleic acids, and the equipment needed to make them, mandatory, and bills including S. 3741, the Biosecurity Modernization and Innovation Act of 2026, were pending in the 119th Congress as of that report. For the scientist at the bench, this shows up as procedure rather than politics: biosafety review before an experiment, and gene orders placed with providers that screen them.

Dear Dr. Whitfield,

I am applying for the Protein Design Scientist position on your binder engineering team. I am a computational biochemist with four years of de novo design experience, and my work has always ended at the bench rather than at a prediction, which is what drew me to a group that runs its own expression and biophysics under the same roof.

In my current postdoc I design small helical binders against cell surface receptors. A typical campaign starts with target analysis and hotspot selection, moves through diffusion-based backbone generation and sequence design, and ends with a filtered panel I order as synthetic genes and test myself. On my last project I took a panel from design to confirmed nanomolar binders in a single round, then solved a cryo-EM structure that matched the designed interface closely enough to guide a second generation with improved thermal stability. I also rebuilt our filtering pipeline so that every variant, score and assay result lives in one queryable database, which cut the time between a design idea and an ordered construct from weeks to days.

I write production quality Python, run jobs across our Slurm cluster, and maintain the internal package our lab uses for scoring and triage. Just as important, I am comfortable saying when a design campaign has failed and why, because the failures are what teach the next specification. I follow institutional biosafety review and order synthetic DNA only through providers that screen orders and customers.

I would welcome the chance to discuss how my design and validation experience fits your enzyme and binder programs, and I can share structures and data from published campaigns at your convenience.

Sincerely, Maya Okonkwo