# Cell Surface Mapping for Drug Design The next wave of targeted drugs comes from three maps of the cell surface: what is there, what sits next to what, and where a drug goes after it binds. All three can now be measured at scale. Source: Frank Gao, "Six Nanometers Apart", *Dimension Research*, Oct 2026.^[Source: [Dimension Research](https://research.dimensioncap.com/p/six-nanometers-apart)] ## The scale of the problem - The membrane is ~5 nm thick. Surface proteins sit ~6 nm apart. - One cell carries 10⁶–10⁷ surface proteins of 2,000–3,000 types. - Abundance spans five orders of magnitude: ~10⁶ copies of HER2 down to <10 copies of a neoantigen. - Approved cancer drugs hit only ~30 surface targets. Most of the surface is unused. ## The three maps | Map | Question | Drug class it unlocks | HER2 example | |---|---|---|---| | **Surfaceome** | What is on the cell, and how much? | Antibodies, ADCs | Trastuzumab: HER2 is abundant | | **Proxiome** | What sits next to what? | Bispecifics that need two targets | Pertuzumab: HER2 pairs with HER3 | | **Deliverome** | Where does the drug go after binding? | ADCs, RNA delivery | Kadcyla, Enhertu: trafficking and release | ```mermaid graph LR S[Surfaceome<br/>what is there] --> P[Proxiome<br/>what is next to what] P --> D[Deliverome<br/>where it goes] D --> M[Safer, more precise drugs] ``` ## Key ideas - **The AND gate.** A bispecific acts only on cells that show both targets. This spares healthy cells that show only one. - **Pair proven with new.** Near-term bet: one validated target + one less-validated target. Example: EGFR × CDCP1. - **Delivery is the bottleneck.** <1% of an ADC dose reaches the tumour. siRNA escapes the endosome only ~1–2% of the time. - **Rank order transfers, magnitude does not.** Across 19 ADCs, lab and animal uptake correlate (ρ 0.82), but animal values run ~27× higher. - **Side effects follow the payload and linker.** Cleavable linkers raise lung disease risk (Enhertu). Other ADCs cause more neutropenia. - **Old tissue is new data.** New proximity methods work on FFPE blocks, which make up >90% of US archived research tissue. ## Who is building this - Cartography Biosciences: annotated atlas of tumour and normal cells to pick target pairs. - InduPro: proximity labelling + graph-based ML. Lilly deal up to ~$950M in milestones. - Nosis Bio: atlas of internalising receptors + AI design. - Deliverome Project: a nonprofit research effort that standardises open delivery data. ## Data lesson - Companies keep their maps proprietary. So open datasets have outsized value. - Target-pair selection is a ranking problem over a huge search space. It needs integrated data on expression, interaction, safety and trafficking. ## Application - [[Drugverse]]: rank candidate target pairs with the knowledge graph. Use OnSIDES and TWOSIDES for payload and combination side effects. Add open delivery datasets as a future source. ---- tags: #boat #deeptech links: [[TechBio MoC]] [[Antibodies]] [[Biologics]] [[Target Identification & Validation]] [[precision medicines]] [[Drug Discovery]] [[Drugverse]]