Protein Representation Learning with Sequence Information Embedding: Does it Always Lead to a Better Performance?

TL;DR

This paper investigates whether embedding amino acid types always improves protein representation learning, demonstrating that a structure-only approach can outperform traditional methods in certain tasks.

Contribution

Proposes ProtLOCA, a local geometry alignment method based solely on amino acid structure, challenging the assumption that amino acid embedding always enhances performance.

Findings

ProtLOCA outperforms existing methods in global structure-matching tasks.

It provides a valid solution for local structure pairing among proteins with different structures.

Embedding amino acid types may not always improve deep learning models for protein analysis.

Abstract

Deep learning has become a crucial tool in studying proteins. While the significance of modeling protein structure has been discussed extensively in the literature, amino acid types are typically included in the input as a default operation for many inference tasks. This study demonstrates with structure alignment task that embedding amino acid types in some cases may not help a deep learning model learn better representation. To this end, we propose ProtLOCA, a local geometry alignment method based solely on amino acid structure representation. The effectiveness of ProtLOCA is examined by a global structure-matching task on protein pairs with an independent test dataset based on CATH labels. Our method outperforms existing sequence- and structure-based representation learning methods by more quickly and accurately matching structurally consistent protein domains. Furthermore, in local structure pairing tasks, ProtLOCA for the first time provides a valid solution to highlight common local structures among proteins with different overall structures but the same function. This suggests a new possibility for using deep learning methods to analyze protein structure to infer function.