RUPEE

RUPEE performs purely geometric protein structure searches to identify structural similarities between predicted or experimental protein structures and known structures, supporting evaluation and refinement of protein structure predictions.

Key Features:

• Purely Geometric Search: Employs a purely geometric approach to identify structural similarities independent of sequence similarity.
• Novel Linear Encoding: Encodes protein structures as a novel linear sequence of residue descriptors to enable effective comparison across diverse conformations.
• Alignment Techniques: Performs Needleman–Wunsch alignments on encoded residue-descriptor sequences to initialize a series of TM-align structure alignments, progressing from TM-align initialized by Needleman–Wunsch to standard TM-align.
• Effective Alignment Normalization: Incorporates alignment normalization at each stage to support both full-length and containment searches and to identify structural motifs within proteins.

Scientific Applications:

• Evaluation of Structure Prediction Methods: Enables validation and refinement of protein structure predictions by identifying database structures most similar to predicted models.
• Assessment of Deep Learning Predictions: Applicable to assessing predictions from deep learning models such as AlphaFold by finding analogous known structures.
• Discovery of Novel Structural Relationships: Detects alignments missed by other tools to uncover novel structural insights and motifs.

Methodology:

Uses a purely geometric search with a novel linear encoding of protein structures as sequences of residue descriptors; performs Needleman–Wunsch alignments on the encoded sequences to initialize TM-align alignments and then standard TM-align; applies alignment normalization at each stage; benchmarked on CASP13 protein structure predictions and compared to mTM-align, SSM, CATHEDRAL, and VAST using RMSD, TM-score, Q-score, and SSAP-score.