SHREC 2025: Protein surface shape retrieval including electrostatic potential

2025-09-16

AI Summary

To address performance degradation in protein surface retrieval caused by data scarcity in certain classes, this paper proposes a multimodal deep learning framework that jointly models molecular surface geometry and electrostatic potential. We introduce a unified surface feature representation that integrates electrostatic potential—a key biophysically grounded descriptor—with 3D geometric structure, thereby overcoming representational limitations inherent to unimodal approaches. Evaluated on a large-scale dataset comprising 11,555 protein surfaces, our method achieves significant improvements in retrieval accuracy, precision, and F1 score; notably, it yields up to a 12.7% relative gain over baseline methods for underrepresented classes. Ablation studies confirm the efficacy of jointly encoding shape and electrostatics. This work establishes a scalable, highly discriminative paradigm for few-shot protein functional annotation and structural alignment.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Representation, semantic annotation, enhancement, enrichments, access and/or integration of a variety of data on the Web
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

This SHREC 2025 track dedicated to protein surface shape retrieval involved 9 participating teams. We evaluated the performance in retrieval of 15 proposed methods on a large dataset of 11,555 protein surfaces with calculated electrostatic potential (a key molecular surface descriptor). The performance in retrieval of the proposed methods was evaluated through different metrics (Accuracy, Balanced accuracy, F1 score, Precision and Recall). The best retrieval performance was achieved by the proposed methods that used the electrostatic potential complementary to molecular surface shape. This observation was also valid for classes with limited data which highlights the importance of taking into account additional molecular surface descriptors.

Problem

• Evaluating protein surface shape retrieval methods
• Assessing electrostatic potential as molecular descriptor
• Testing performance on large protein dataset

Innovation

• Combined electrostatic potential with surface shape
• Evaluated 15 methods on 11,555 protein surfaces
• Used complementary descriptors for improved retrieval