Mutations are often thought to destabilize protein structures. However, many proteins are naturally unstructured and the effect of mutations on such proteins especially in the absence of selection, remains unclear. Here, we develop a computational model to study the effects of mutations on structural disorder in both random and natural sequences, including those derived from evolutionary conserved proteins, and evolutionarily novel de novo proteins. We find that while structured proteins tend to lose structure, unstructured proteins exhibit the opposite trend, becoming more structured in the absence of directional selection. This bidirectional dynamics is robust to mutation biases, genetic code structure, and sequence origin, suggesting that it arises from the topology of the sequence-structure landscape rather than intrinsic directional mechanisms. Our results are consistent with a diffuse landscape in which structured and disordered sequences are interspersed throughout sequence space. These findings suggest that neutral evolution can result in structure formation and may facilitate the early structural evolution of de novo proteins prior to strong selection.
Target-Side Paraphrase Augmentation for Sign Language Translation with Large Language Models
arXiv:2605.31393v1 Announce Type: cross Abstract: Sign language translation (SLT) remains constrained by limited paired sign-video/text corpora and heavy-tailed target vocabularies. We study target-side augmentation in



