Vision transformer autoencoders captures local and non-local features in brain imaging to reveal novel genetic associations
Samia R Islam, Tian Xia, Wei He, Ziqian Xie, Degui Zhi · 2026
Read Vision transformer autoencoders captures local and non-local features in brain imaging to reveal novel genetic associations by Samia R Islam, Tian Xia, Wei He, Ziqian Xie, Degui Zhi (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/pmc-13635212.txt.
Description
Brain scans can reveal new genetic links when a machine-learning model learns their patterns without preset measurements, Islam and colleagues argue. Linking genetic variation to brain structure usually relies on hand-chosen features, which may miss subtle patterns. The team trained a vision transformer autoencoder on magnetic resonance images from more than six thousand UK Biobank volunteers. This model compresses each scan into a compact numerical summary, which they then used in genetic studies of nearly twenty-three thousand people.
The approach uncovered sixty-three genetic locations, including twenty-four missed by an earlier convolutional method, and interpretation showed the model captured both local detail and whole-brain patterns such as left-right symmetry. Those broad patterns, picked up through the model's attention mechanism, set it apart from the earlier approach. Islam and colleagues conclude that transformer-based image summaries offer a richer route to gene discovery.
Catalog record
| CREATOR | Samia R Islam, Tian Xia, Wei He, Ziqian Xie, Degui Zhi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |