Using protein blocks to build custom fragment libraries from protein structures

Abstract

The remarkable structural diversity of modern proteins reflects millions of years of evolution, during which sequence space has expanded while many structural features remain conserved. This conservation is evident not only among homologous proteins but also in the recurrence of supersecondary motifs across unrelated proteins, underscoring the abundance and robustness of these structural units. Here, we present a novel pipeline for generating customized protein fragment libraries using protein blocks (PBs)—a structural alphabet that encodes local backbone conformations. Our method efficiently extracts structurally similar fragments from a curated, non-redundant protein structure database by converting three-dimensional structures into one-dimensional PB sequences. By integrating predicted PB sequences with the PB-ALIGN and PB-kPRED tools, our approach identifies relevant fragments independently of sequence homology. Fragment quality is further assessed using a new scoring function that combines secondary structure similarity and PB alignment metrics. The resulting libraries contain fragments of at least seven PBs (11 amino acid residues), covering over 70 % of the local backbone structure. Our results demonstrate that PBs enable efficient mining of high-quality structural fragments from diverse protein spaces, including proteins with disordered regions. The pipeline is accessible as an online tool (PB-Frag, http://pbpred-us2b.univ-nantes.fr/pbfrag).