Beyond profiles: supervised repeat annotation using protein embeddings
Year: 2026
Type: preprint
Abstract: Protein repeats give rise to diverse protein structures and functions, yet their sequence-based detection is often challenging because individual repeat units can diverge substantially in sequence while retaining similar structures. Most sensitive approaches rely on sequence information directly represented by alignments or derived profiles, thereby limiting flexibility and scalability. Here, we introduce TREAD ( T ransfer learning-based RE peat A nnotation using Protein Embe D dings), a supervised framework that reformulates repeat detection as a residue-level annotation problem and operates directly on embeddings from protein language models. Instead of constructing explicit probabilistic profiles, TREAD learns repeat-specific features implicitly, enabling residue-resolved scoring and flexible repeat segment extraction. Across complementary benchmarks based on RepeatsDB and Pfam, TREAD consistently matches or outperforms the widely used profile-based tool HMMER, particularly in low-data and high-divergence settings. The model exhibits robustness to score thresholding and demonstrates better generalization to independent test sets, including those containing remotely related families. To illustrate the practical utility of TREAD, we applied it to survey β-propeller proteins across the AlphaFold Database and representative proteomes to generate a comprehensive census of this fold. This analysis highlights extensive propeller diversity, identifies lineage-specific expansion patterns across the tree of life, and suggests previously unrecognized relationships between propellers and other repeat folds. Together, TREAD provides a flexible and scalable alternative to profile-based repeat annotation and establishes a general motif-centric framework for protein sequence annotation. Author Summary Many proteins are assembled from repeated building blocks. These repeats have helped evolution generate a remarkable variety of structures and functions. Yet sensitive detection of individual repeats has long been challenging because their sequences can diverge substantially even when their structures remain similar. Conventional methods often rely on comparing large collections of related sequences to find these patterns. We developed TREAD, a computational method that learns from annotated examples to recognize protein repeats using information captured by protein language models. Rather than building conventional sequence profiles, TREAD identifies repeat units and their boundaries directly from these learned protein representations. In systematic benchmarks, it matched or outperformed HMMER, a widely used profile-based method, particularly when few examples were available or repeat sequences were highly divergent. We then used TREAD to survey β-propellers, proteins assembled from repeated blade-like units, across millions of protein sequences. We uncovered substantial unexplored diversity, differences in their abundance and associated functions across organisms, and unexpected similarities to a previously unclassified barrel-like repeat fold. Together, our work provides a flexible approach to detecting protein repeats and shows how their systematic annotation can reveal new aspects of protein diversity and evolution.
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7162030659