I'm a senior CS student at Yale, graduating in May. I took my entire junior year off to work in industry. I was the first hire of Palantir's 2025 off-cycle program, joining the hospital operations team as an FDE. I traveled to hospitals to round with nurses and developed a mobile app now used for nurse scheduling in over 100 hospitals. I also spent part of my year off at Amazon's Moonshot lab, developing a novel algorithm (currently patent pending) as the sole scientist under one of their "founders in residence." I took the project from zero to a CEO demo in three months, after which it received eight-figure funding under Amazon Special Projects. In July I went onsite to implement a pilot for the first customer engagement and have been advising since.
Last month I was talking with a researcher at the Doudna CRISPR lab at Berkeley. I asked him how difficult it would be for someone with his expertise to create a global pandemic equivalent to COVID, and he replied that it would be relatively easy. I found this deeply disturbing, and started reading whatever papers I could. I left for a backpacking trip in Vietnam a few days later, but I had a Kindle with me and read a cellular biology textbook to better understand the language. The more I learned, the clearer it became that currently there is no real defense against these types of threats.
I believe Valthos is by far the most talent dense team working on this, and the most likely to succeed. Starting with an ISR product is the right call. Without threat detection our government is flying blind.
It was clear to me that the fastest way to gain a basic understanding of synthetic biology and current frontier capabilities was to get hands-on, not just read a textbook. I designed a computational experiment that was novel, but simple enough that it could be verified in vivo safely in a dorm. I gave myself a two week time budget to execute it.
Several mutations in the rpsL gene are known to cause antibiotic resistance in lab strains of E. coli, which makes edits easy to verify by plating on an antibiotic medium. However, many possible mutations at this locus have not been tested. I used AlphaFold to compute structural similarity between novel and known mutations, then ranked those most likely to succeed. To validate the testing pipeline, I purchased my own oligos and used CRISPR-Cas9 to edit a known mutation in living E. coli. The same pipeline could be used to test the novel predictions, but I stopped short of that for safety.
The bacteria used for this experiment were a non-pathogenic, BSL-1 (Biosafety Level 1) lab strain that can't survive outside of rich media. I did not make any novel genetic edits, only a mutation previously explored in the literature. Experiments were conducted in a still air box while wearing PPE. Plates and liquid cultures were treated with bleach before disposal.
Small ribosomal subunit protein uS12 is a protein coded by the rpsL gene that makes up part of E. coli's ribosomal complex. At least 19 substitutions or deletions in 11 codons of the rpsL gene are known to promote streptomycin resistance, dependence, or pseudodependence. K43 (lysine at position 43) in particular is a known hotspot for resistance mutations. K43R, K43T, and K43N have all been shown to introduce resistance in the DH5-alpha lab strain. The remaining possible codon substitutions have not been tested to my knowledge.
I ran AlphaFold 3 on all possible substitutions at K43 and compared structurally to K43T. All mutants keep essentially the same S12 fold, so the useful differences are local around residue 43. Based on likely functional similarity, K43S is the best candidate. Further analysis looking at the ribosome complex supported this assessment.
The in-depth analysis is detailed in an accompanying Jupyter notebook.
Performing the actual substitution using CRISPR-cas9 requires the cas9 protein, gRNA, and template DNA. For cas9 I used the "pCas" plasmid, one of the most widely used CRISPR tools for E. coli genome editing. For the gRNA, I used a pUC-backbone plasmid available for purchase online which targets the rpsL locus adjacent to K43.
For template DNA, I ordered a custom 58-mer single-stranded oligo that served as the template DNA. This oligo is identical to the wild-type rpsL sequence except for a single A-to-C transversion that converts the K43 codon from lysine to threonine (K43T). This is the known mutation I used to validate the pipeline.
I purchased a still-air box to minimize contamination. I kept a spray bottle of isopropyl alcohol both inside and outside the box to sanitize items entering/exiting. For incubation I used a chicken egg incubator as a low-cost alternative to a lab incubator. I bought a cheap make-up fridge to store plates and samples. I verified with a thermometer that it keeps 4°C consistently.
I prepared LB agar and selective plates (LB/Strep/Kan/Arab), rehydrated freeze-dried DNA and bacteria, and streaked DH5-alpha onto an LB plate for overnight growth at 37°C.
I prepared competent cells with a CaCl₂/PEG transformation buffer, then co-transformed with the pCas9 plasmid, gRNA plasmid, and my custom K43T template oligo. After heat shock and a recovery incubation, I plated onto selective media and incubated at 30°C for 48 hours.
Colonies appeared on the selective plates, confirming successful editing of rpsL.
As a software engineer, I had to adjust to the fact that biology is extremely non-deterministic and a pain to debug. I kept a record of every used plate and its condition in a spreadsheet. After transformation, I streaked out some bacteria onto a non-selective medium (breakpoints, essentially) to distinguish between failed edits and dead cells.
Thank you for taking the time to review my project. I have come to strongly believe that bioweapons will soon become the "minimum viable" WMD. They are much easier to produce than nuclear weapons, easier to deliver than chemical weapons or dirty bombs, and their proliferation could rival nuclear weapons as a strategic deterrent. I find it likely that rogue states will at some point adopt them as an alternative to a nuclear arsenal. The risk of states and non-state actors pursuing such weapons when no countermeasures exist are disturbing. I would love to discuss Valthos and my project further.