CAPSUL: A Comprehensive Human Protein Benchmark for Subcellular Localization

2026/03/19 by Yicheng Hu, Xinyu Lin, Shulin Li +4 · 1 voice
Biochemistry, Genetics and Molecular Biology · Computer Science · #Benchmark (surveying) #Bioinformatics and Genomic Networks #Bridging (networking) #Cell Image Analysis Techniques #Identification (biology) #Interpretability #Machine Learning in Bioinformatics #Protein subcellular localization prediction #Subcellular localization #Task (project management) #cs.AI #cs.CE #q-bio.QM

paper · pdf · open access · doi:10.48550/arxiv.2603.18571

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2026/03/19 · arxiv published 2026/03/19 · arxiv updated 2026/03/19 · openalex created_date 2026/03/21 · openalex updated_date 2026/07/28

Abstract

Subcellular localization is a crucial biological task for drug target identification and function annotation. Although it has been biologically realized that subcellular localization is closely associated with protein structure, no existing dataset offers comprehensive 3D structural information with detailed subcellular localization annotations, thus severely hindering the application of promising structure-based models on this task. To address this gap, we introduce a new benchmark called CAPSUL, a Comprehensive humAn Protein benchmark for SUbcellular Localization. It features a dataset that integrates diverse 3D structural representations with fine-grained subcellular localization annotations carefully curated by domain experts. We evaluate this benchmark using a variety of state-of-the-art sequence-based and structure-based models, showcasing the importance of involving structural features in this task. Furthermore, we explore reweighting and single-label classification strategies to facilitate future investigation on structure-based methods for this task. Lastly, we showcase the powerful interpretability of structure-based methods through a case study on the Golgi apparatus, where we discover a decisive localization pattern α-helix from attention mechanisms, demonstrating the potential for bridging the gap with intuitive biological interpretability and paving the way for data-driven discoveries in cell biology.

Citations

• An AI/ML-modeled structural atlas of the human protein interactome with functional and cancer-focused networks
• Boltz-1 Democratizing Biomolecular Interaction Modeling
• Mamba: Linear-Time Sequence Modeling with Selective State Spaces
• AlphaFold Protein Structure Database in 2024: providing structure coverage for over 214 million protein sequences
• De novo design of protein structure and function with RFdiffusion
• Directional diffusion models for graph representation learning
• Towards Understanding the Effect of Pretraining Label Granularity
• Large language models generate functional protein sequences across diverse families
• Evolutionary-scale prediction of atomic-level protein structure with a language model
• Learning inverse folding from millions of predicted structures
• PEER: A Comprehensive and Multi-Task Benchmark for Protein Sequence Understanding
• Robust deep learning–based protein sequence design using ProteinMPNN
• Recipe for a General, Powerful, Scalable Graph Transformer
• Protein Representation Learning by Geometric Structure Pretraining
• Fast and accurate protein structure search with Foldseek
• Pre-training Co-evolutionary Protein Representation via A Pairwise Masked Language Model
• Highly accurate protein structure prediction with AlphaFold
• Is Transfer Learning Necessary for Protein Landscape Prediction?
• Learning from Protein Structure with Geometric Vector Perceptrons
• Evaluating Protein Transfer Learning with TAPE
• Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss
• Class-Balanced Loss Based on Effective Number of Samples
• A subcellular map of the human proteome

Discussions

• CAPSUL: A Comprehensive Human Protein Benchmark for Subcellular... arxiv.org/abs/2603.1... --- #proteomics #prot-preprint [bsky, 1 points, 0 comments]

Related

• Single-cell Subcellular Protein Localisation Using Novel Ensembles of Diverse Deep Architectures
• HPSLPred: An Ensemble Multi-label Classifier for Human Protein Subcellular Location Prediction with Imbalanced Source
• Extracting Cellular Location of Human Proteins Using Deep Learning
• Learning multi-scale functional representations of proteins from single-cell microscopy data
• A subcellular map of the human proteome