Read by Interest
- Which antibodies to combineCoverage across 525 real viruses; try it in the Coverage Explorer.
- What the sugar coat hidesThe full-density shield in 3D; also the standalone viewer.
- How the virus escapesWhy sequencing depth hides independent events: Sampling Blind Spot and Origin Network.
- How the work was doneThe AI-assisted workflow, methods, and caveats.
- Biosecurity and data sharingPublic health contribution and dual-use considerations.
Interactive Companions
Virion Portrait
A mature HIV particle, cut in half and drawn to scale
Shows how few Env spikes sit on a virion and how far apart they are, alongside the real 5FYJ spike with a full sugar coat and an antibody for scale.
Coverage Explorer
Which of 525 real viruses does an antibody set reach?
Switch antibodies on and off and see how many viruses the set still neutralizes, with a check against the project's standing rule for choosing combinations.
Full-Density Glycan Shield
Complete Man9 glycans built onto every site of the 5FYJ trimer
A crystal structure shows only the first few sugars of each glycan. This 3D viewer builds the full shield so you can see how much of each antibody footprint it hides. It is a snapshot ensemble of a dense case, not a measurement of the real shield.
The Sampling Blind Spot
Why sequencing depth hides most independent escape events
Watch a virus population evolve resistance under antibody pressure, then see most of that resistance disappear once realistic sequencing depth is applied. The page uses a simplified illustrative model and gives the real-scale reference numbers beside it.
Origin Network
One participant's escape lineages, drawn out
Takes one participant from the 10-1074 monotherapy cohort through the source paper's own ancestor-matching code and draws each viral haplotype and each independent origin of escape, next to a simulated replicate.
About This Project
This is independent computational work, separate from my postdoctoral research. The report and companions are self-contained pages: each states its data sources, methods, and caveats, and cites the primary literature it draws on.
The reports were developed with AI-assisted tooling. The analyses are computational and exploratory; they are not experimental results and should not be read as clinical guidance.