Tiny transformers trained from scratch classify six protein families with high accuracy and their gradient saliency maps align with known catalytic motifs, though the claim that this demonstrates biological intelligence is not supported by the evidence.
Accurate prediction of protein structures and interactions using a three-track neural network
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Do Protein Transformers Have Biological Intelligence?
Tiny transformers trained from scratch classify six protein families with high accuracy and their gradient saliency maps align with known catalytic motifs, though the claim that this demonstrates biological intelligence is not supported by the evidence.