What topological and geometric structure do biological foundation models learn? Evidence from 141 hypotheses
Ihor Kendiukhov · 2026
Read What topological and geometric structure do biological foundation models learn? Evidence from 141 hypotheses by Ihor Kendiukhov (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0344826.txt.
Description
Ihor Kendiukhov reports an unusual experiment in this 2026 PLOS ONE paper. An automated loop of AI programs proposed hypotheses. It then tested one hundred and forty-one of them. The subject was what AI models of cells actually learn. These models read single-cell gene activity. They are widely used in biology. The question was simple. Does the geometry inside them reflect real biology? Or is it just a training artefact?
Several patterns held up well. Gene neighbourhoods had genuine mathematical shape. Curved distance measures found regulatory gene pairs better. Two independently trained models agreed on the overall shape of gene space. Yet the strictest checks told a narrower story. Solid signal concentrated in immune tissue. The paper, negatives included, calibrates how much trust these models have earned.
Catalog record
| CREATOR | Ihor Kendiukhov |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54359 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |