STELLAR Enables Structural Prediction of Highly Flexible Long-Chain Biopolymers via a Novel Fragment-Based Workflow
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This Zenodo repository contains the following data: Raw data obtained from three docking cases involving protein complexes with 4-, 5-, 6-, and 7-fragment peptides, as well as polysaccharides (Frag4.zip, Frag5.zip, Frag6.zip, Frag7.zip, and Polysaccharides.zip). Input data required to run STELLAR in the benchmark proposed in the article (Propedia_benchmark.zip and Polysaccharides_benchmark.zip). Raw data for the interpretable machine learning analysis of the 8YP8 case (Sibila_8YP8.zip). Execution times for cases 1J4Q (4 fragments), 2ND1 (5 fragments), 1BXL (6 fragments), and 3R7G (7 fragments) for a single combination run (execution_time_logs.zip). The remaining cases and additional data generated by STELLAR are available from the authors upon reasonable request. Summary Docking methods have improved significantly with AI optimization and systematic exploration of parameter configurations within docking software. However, docking large molecules like peptides or carbohydrates remains challenging. Existing tools struggle with accuracy and efficiency for peptides of larger lengths, and generative Deep Learning and Machine Learning approaches still lack robust ranking metrics and the performances could be suboptimal. To address this, we developed STELLAR (Score-Tuning for Efficient ranking of Large Ligands using an Accurate and Refined docking configuration), a workflow able to process docking for peptides longer than 10 residues. STELLAR implements a fragment-based docking (FBD) strategy, which decomposes peptides into smaller units for individual docking and subsequent recomposition into full-length structures. Built on the MetaScreener platform (https://github.com/bio-hpc/metascreener), STELLAR includes optimized algorithms for handling large peptides. The pipeline also integrates structural optimization steps using tools such as RDKit, PyMOL and GROMACS, ensuring physically realistic poses. STELLAR achieved low RMSD values in the validation conducted using benchmark complexes from Propedia and PDB. It outperforms several state-of-the-art tools such as HPepDock, DiffDock, and CABS-Dock across a range of peptide lengths (up to 25 residues tested). It scales linearly in computing time (increasing with every three amino acids), runs efficiently on CPUs, avoids reliance on stochastic approximations, obtains several metrics to rank different ligands, and reduces computing time compared to other similar tools. Its design also supports high-throughput screening for linear polymer-protein interactions by efficiently managing conformational flexibility.
Fields of Study & Subjects
Computational Drug Discovery Methods
Protein Structure and Dynamics
vaccines and immunoinformatics approaches
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 24 March 2026
- Language
- EN