UC San Diego
B.S. Biochemistry and Cell Biology, 2018
Huffaker Lab, UCSD
USDA-ARS area-wide pest management program
The invasive sugarcane aphid in grain sorghum
Aphids land and attach.
Most plants suffer. A few resist.
What makes the survivors different?
The molecular interactions between aphid and host.
Searching for genetic loci associated with resistance.
We screened hundreds of genetically diverse sorghum varieties for their resistance to sugarcane aphids and performed a genome-wide association study (GWAS) to identify genetic variants associated with resistance.
We identified a strong association near the transcription factor SbWRKY86.
We then used gene-expression and functional experiments in other plant systems to show that increased expression of this gene reduced aphid populations.
Singular Genomics
Seven years. One company's full lifecycle.
Story one
Building the G4 flow cell.
Start with nanoimprinted glass.
Standard glass works, but random template seeding is Poisson-limited: as occupancy rises, so does the chance of multiple templates landing on one feature.
Nanoimprinted, patterned flow cells with controlled seeding chemistry target one template per feature, minimizing empty and polyclonal features.
Which glass, which vendor, which material, and what stays compatible with the workflow?
Fill the wells with polymer.
- Polymer chemistry & architectureRAFT · PEG · linear vs. branched / tree-like
- Degree of polymerizationTuning polymer chain length
- Surface attachmentGrafting-to vs. grafting-from
- Surface & substrate compatibilityM-type · Ormocomp · BD
- Optimize for the biologyPrimer density · clustering · cbPCR performance
Functionalize the surface with primers.
- Attachment chemistryDBCO-mediated coupling / click chemistry
- Primer densityDegree of labeling · surface density
- Primer architectureSpacer length · PolyT · ss vs. ds presentation
- Coupling conditionsConcentration · time · temperature
Seed the wells with DNA.
- Template concentration & seeding densityControl occupancy of individual nanowells
- Seeding chemistrySalt · buffer · temperature
- Library propertiesLibrary type · concentration · denaturation
- Loading conditionsTime · wetting · fluidics
- Maximize usable wellsHigh occupancy · minimize mixed clusters
Clonally amplify each template.
- Amplification chemistryPolymerase · Betaine · Mg²⁺ · buffers
- Thermal conditionsTemperature · extension time · cycle number
- Reaction environmentpH · salts · additives
- Amplification efficiencyCluster intensity · density · uniformity
- Optimize for sequencingHigh-quality, spatially confined clonal clusters
- Optimize for amplificationAccessibility · hybridization · cbPCR performance
From above, the clusters are the dots from the start.
Then prove an instrument can do it.
G4X: same toolkit, new problem.
Tissue on glass that stays put.
Co-inventor, U.S. Patent: Solid Supports Useful for Tissue Adherence
One workflow, every generation.
From a benchtop assay to a fully integrated sequencing workflow.
Fluidics
Tubing · valves · port mapping
Flow rates · washes
Thermal control
Temperature · incubation
Cycling · reagent stability
Automation
Template seeding · cbPCR
Cluster processing
Troubleshooting
Chemistry × fluidics
Hardware × software
Back from the Fourth of July break.
We powered the instruments back on and ran system diagnostics.
Nearly 100% failing.
Where does the failure live?
Bisect the workflow.
Every step can run by hand or on the instrument.
Move one block at a time and watch the result.
Leading theory: the bleach.
A bleach formulation change took effect about two weeks before the break.
Kit timing meant the new formulation first ran right after the shutdown.
Bleach runs through first.
Working hypothesis: it damages the manifold.
Then the DNA follows.
Working hypothesis: it sticks in the manifold, or breaks down on the way.
Almost nothing reaches the flow cell.
The fix.
- Rinse the system thoroughly
- Swap the affected parts
- Bleach 100x lower
Something had broken across nearly the entire fleet.
~500Mexpected reads / flow cell
versus
<1Mobserved
The same failure was appearing across independent instruments.
Run the workflow manually
WORKS
Run the workflow onboard
FAILS
The biology worked. Something about integration didn't.
Steps 1 to 4 on the instrument. It works.
Steps 5 and 6 on the instrument. Still works.
Step 7 alone on the instrument. It fails.
Story two
Assay Troubleshooting.
Biology and engineering share failure modes.
The exact mechanism was never fully solved. It wasn't critical to the business goal at the time, so we didn't pursue it further.
Story three
Multiplexed Protein Detection.
Reading protein in situ.
- Antibody-oligo conjugate binds its epitope
- Fixed in place
- A near-complete DNA circle hybridizes
- Ligation closes the circle
- Rolling circle amplification
- One channel lights up
Color tells us target. Timing adds capacity.
× 6 primers = 144 theoretical slots, toward the ~120-target goal
More proteins. The same hardware.
Higher-plex panels for applications such as ADC evaluation.
A delayed signal can look like another protein.
When timing encodes identity, carryover matters.
Early tests showed substantial carryover: signal appearing in a later slot could be mistaken for another target.
A step back to fundamentals.
After extensive troubleshooting, we revisited a fundamental design choice: the DNA backbone.
Changing the backbone revealed a key contributor to the unwanted signal.
Keep the dim cells visible.
Later read positions made some weaker signals harder to detect.
More targets. Reliable biological detail.
Refining the readout improved signal, with further work needed to reach the full multiplexing goal.
Cycle 1
GTCA
GTCOff (dark): not a target
↓
Intended signal Delayed echo
Schematic
More target capacity
24 toward a ~120-target goal
Clearer target separation
Smaller delayed echo
Weak signals more visible
Further work still needed
Outside the lab.
Photography
Eight years running a photo business.
Surf
Snowboard
Scuba
Cook
Sarthak Duggal
Why Exthymic.
- A small team doing R&D
- Cell therapy
- Surface and consumables work
How I built this.
- Claude Code
- AI-generated flow cell visuals
- GitHub
- Deployed to my own subdomain
- 11 commits