Refining Structure-Based Search in AlphaFold DB: Including confidence metrics into AlphaFind

Protein structure prediction has seen remarkable advancements with the advent of AlphaFold DB, yet efficient structure-based searches remain a challenge. Within AlphaFind, a novel tool for structure-based similarity search in AlphaFold DB, we seek to enhance protein comparison by integrating multiple confidence metrics such as predicted local distance difference test (pLDDT) and predicted aligned error (PAE). The proposed method refines search results through a two-step process: fast vector-based pre-filtering followed by precise structure alignment using TM-score. Additionally, high-confidence regions are extracted based on stability criteria, allowing for a more reliable comparison of protein structures while mitigating the impact of low-confidence residues. This approach enables searches across multiple levels, from individual fragments to full protein chains, broadening the scope of potential structural matches. Future developments aim to further integrate data-driven filtering methods and molecular visualization tools, improving the accuracy and usability of structure-based searches.