VAEBAC: A representation-learning framework for proteome-scale prediction of amyloidogenic proteins and functional...

Nucleic-acid-binding proteins (NABPs) govern transcription, replication and genome organization, yet the distribution of aggregation susceptibility within regulatory proteomes remains unresolved. Here, we systematically examine amyloidogenic potential across the nucleic-acid-binding proteomes of four phylogenetically diverse organisms-Escherichia coli, Bacillus subtilis, Saccharomyces cerevisiae and Homo sapiens-spanning prokaryotes and eukaryotes, to determine whether aggregation-prone sequence architectures are selectively constrained within transcriptional networks.

Using VAEBAC, a representation-learning framework that integrates large-scale amyloid-family sequences with experimentally validated annotations, we performed proteome-scale prediction of amyloidogenicity across nucleic-acid-binding proteins in each organism. Predicted amyloidogenic proteins were consistently and significantly enriched in Gene Ontology categories associated with regulation of DNA-templated transcription, RNA biosynthesis, and gene expression across all four organisms, a pattern validated using three independent annotation tools. Fisher's exact test confirmed significant enrichment of amyloidogenic NABPs among transcription regulators in eukaryotic proteomes. Residue-level mapping demonstrated spatial separation between predicted aggregation-prone regions and DNA-binding interfaces, suggesting structural compatibility between regulatory function and conditional aggregation. Experimental validation confirmed fibrillar assembly of selected proteins. These findings support a conserved model in which aggregation susceptibility is functionally stratified across transcriptional regulatory tiers rather than uniformly suppressed within NABPs, a principle conserved across 3.5 billion years of evolution.

VAEBAC is freely accessible at https://www.vaebac.com.

Supplementary data are available at Bioinformatics online.