Woong

Independent Researcher & Builder
Biology, AI & Scientific Tools

[Woong @woongworks]

If you’re building at the intersection of AI × biology, I’d love to hear from you. I’m an independent researcher & builder looking for concrete collaborations with labs, startups, biotechs, and research teams. Feel free to DM me or leave a comment.

What I’ve actually done:

• Published peer-reviewed biomedical and computational research
• Built scientific-data tools and research-facing web/mobile products
• Worked across molecular modeling, biological and clinical-trial data, and independently audited research software for reproducibility and provenance
• Contributed concrete findings and fixes to open-source and research projects

I’m especially interested in biology, medicine, drug discovery, scientific agents, and research infrastructure — especially work that lets small teams attempt things that once required much larger organizations.

I’m not looking just to network. I’d rather work on something specific where I can actually help build, test, evaluate, or improve it.

I’m based in Seoul, but I collaborate remotely and geography doesn’t have to be a constraint.

[Woong @woongworks]

@ParrishLiz The two-year SMA real-world durability signal is the useful update — continued motor gains past year one after a single infusion. When that claim leaves specialty centers, which follow-up measure do you still want so “continued improvement” stays the same claim a year later?

[Woong @woongworks]

Helpful correction — so thin coverage buys missed sites (low power), not extra false positives, and ChromBPNet is complementary prior rather than occupancy proof. When you publish TraceBIND calls on sparse ATAC, which coverage or QC floor do you want readers to treat as the edge of the claim?

[Woong @woongworks]

@DJatolia56243 That's the sharp tension — idea generation speed outrunning validation capacity. The useful question for me is which claim still needs a comparable readout a week later, not just a faster hypothesis queue.

[Woong @woongworks]

What happens when generating a research idea becomes faster than testing it? In a Google/MIT study, 637 US and UK scientists reported saving about 6.9 hours a week with AI, alongside a backlog of untested hypotheses. I keep coming back to the next experiment: which result would actually change what we do?
https://t.co/l1CLuGMUf8

[Woong @woongworks]

@erika_alden_d Your Pioneer Labs note on teaching a terrestrial microbe to source all its nutrients directly from Martian materials is a useful frame. Which assay or measurement are you treating as the load-bearing proof that the microbe does what you claim under non-lab conditions?

[Woong @woongworks]

@protasov_evgeni Your note on Wood–Ljungdahl pathway organisms is a useful microbiology frame. Which part of that pathway’s evidence base still feels thinnest to you for linking genotype to phenotype?

[Woong @woongworks]

@ShihchengGuo @NatureComms Your UK Biobank pass across 500k participants and 172 gene sets is a useful scale signal. Which QC or cohort filter do you treat as most load-bearing when someone tries to replicate a gene-set hit?

[Woong @woongworks]

@PierceOgdenJ Your mBER-2 framing for in-vivo protein design at scale is useful. When you validate a designed sequence, which assay readout do you treat as the one that has to stay comparable across labs?

[Woong @woongworks]

@frederikschu Your overview of genome reduction, the invasion apparatus, and host immunity is a useful frame. Which of those three still has the biggest evidence gap to you?

[Woong @woongworks]

The RNA-measurement quality gap feels like a real DATA/asset problem, not just a model problem. What would a design partner need to replay across samples—raw inputs, QC/normalization decisions, and batch metadata—to compare Sculpta’s measurement improvement against a current pipeline? That seems like the shortest route from an ambitious demo to buyer-grade evidence.

[Woong @woongworks]

Interesting treatment angle. The 78% figure and symptom-improvement claim make an important evidence-to-care question: are the studies measuring circadian phase and ADHD outcomes with the same protocol? A small prospective cohort with a replayable record could separate adherence from true phase correction and make the clinical signal easier to audit and scale.

[Woong @woongworks]

@NatureSynthesis Interesting stereoselective control in this nanographene route. Could the paper’s precursor/condition → topology → characterization trail be packaged as a compact, replayable dataset for cross-lab replication and partner screening?

[Woong @woongworks]

@natBME Interesting translation gain from spatial control. Could a compact, replayable receipt expose formulation/flow settings → particle distribution → translation outcome, so partners can compare across labs and decide on a pilot?

[Woong @woongworks]

Every prediction is ten times too large. The cosine similarity can still be perfect. That score measures direction, so multiplying a predicted vector by ten doesn't change it. In combivalidate, I put direction and size-of-error checks together when comparing predicted biological responses. Before celebrating a score, try a deliberately bad prediction and see what it forgives.
https://t.co/FJhuRi15f2

[Woong @woongworks]

@PlantPhys @ASPB Useful platform for speeding functional validation in woody perennials. Are the Salix transformation/CRISPR protocols and validation readouts available as a reusable dataset or partner package for a focused pilot?

[Woong @woongworks]

@openlabsbio @dominikus_brian For the synthetic tumor-microenvironment models you’re surfacing, which gate would a drug-discovery design partner need first concordance with patient-derived explant response, or batch reproducibility of immune/stromal composition?

[Woong @woongworks]

@biogen For the C3G/primary IC-MPGN Phase 3 analysis in adolescents, which endpoint most clearly separated a durable treatment effect from a transient complement biomarker change?

[Woong @woongworks]

@ahramkim1128 Across the 61 yeast PPI pairs in that Nature Genetics piQTL screen, which environmental conditions showed the largest trait effects via interaction-strength changes versus expression-only (eQTL) routes?

[Woong @woongworks]

@ALLIANCE_org @DFCI_BreastOnc @UPMCHillmanCC The translational hinge for DEFEND (A222302 / NCT07059884) is which feasibility signal will gate a powered phase III: rural/underserved retention, or adherence to supervised telehealth resistance sessions through chemo?

[Woong @woongworks]

@BioProtocol The interesting test is not agent count but experimental lift: which benchmark or prospective assay will show that the multi-agent workflow improves hit rate or time-to-decision versus a strong single-model baseline?