Abstract:Accurate prediction of enzyme function is crucial for elucidating biological mechanisms and driving innovation across various sectors. Existing deep learning methods tend to rely solely on either sequence data or structural data and predict the EC number as a whole, neglecting the intrinsic hierarchical structure of EC numbers. To address these limitations, we introduce MAPred, a novel multi-modality and multi-scale model designed to autoregressively predict the EC number of proteins. MAPred integrates both the primary amino acid sequence and the 3D tokens of proteins, employing a dual-pathway approach to capture comprehensive protein characteristics and essential local functional sites. Additionally, MAPred utilizes an autoregressive prediction network to sequentially predict the digits of the EC number, leveraging the hierarchical organization of EC classifications. Evaluations on benchmark datasets, including New-392, Price, and New-815, demonstrate that our method outperforms existing models, marking a significant advance in the reliability and granularity of protein function prediction within bioinformatics.

What problem does this paper attempt to address?

The paper aims to address several key issues in enzyme function prediction, particularly the accurate prediction of Enzyme Commission (EC) numbers. The main problems the paper attempts to solve are as follows: 1. **Combining Sequence and Structure Information**: Existing deep learning methods often rely solely on either protein sequence data or structure data for prediction, neglecting the more comprehensive information that could be obtained by combining both. The proposed method (MAPred) integrates the primary sequence of amino acids and the three-dimensional structure information of proteins (represented by 3Di tokens) to achieve a more comprehensive protein characterization. 2. **Multimodal and Multiscale Fusion**: To capture both the global properties of proteins and the features of local functional sites, the paper proposes a dual-path feature extraction network, including a global feature extraction path and a local feature extraction path. This allows the model to understand protein characteristics from different perspectives. 3. **Utilizing Autoregressive Prediction**: Existing methods typically predict the EC number as a whole, ignoring the inherent hierarchical structure of the EC number. MAPred adopts an autoregressive prediction strategy, sequentially predicting each digit of the EC number, thereby better leveraging the hierarchical organization of the EC number. 4. **Improving Prediction Accuracy**: Through the aforementioned innovations, the paper aims to enhance the reliability and granularity of protein function prediction, particularly excelling in the prediction of small sample sizes and low-frequency EC numbers. In summary, this research aims to improve enzyme function prediction, especially the accuracy of EC number prediction, by introducing a multimodal, multiscale autoregressive prediction framework. This approach is expected to advance the understanding of enzyme catalytic mechanisms, substrate specificity, and potential industrial applications in the field of bioinformatics.

Autoregressive Enzyme Function Prediction with Multi-scale Multi-modality Fusion

Enzyme Commission Number Prediction and Benchmarking with Hierarchical Dual-core Multitask Learning Framework

FEDKEA: Enzyme function prediction with a large pretrained protein language model and distance-weighted k-nearest neighbor

ECRECer: Enzyme Commission Number Recommendation and Benchmarking based on Multiagent Dual-core Learning

Accurately predicting enzyme functions through geometric graph learning on ESMFold-predicted structures

Deep learning enables high-quality and high-throughput prediction of enzyme commission numbers

Predicting Enzyme Functions Using Contrastive Learning with Hierarchical Enzyme Structure Information

An Interpretable Double-Scale Attention Model for Enzyme Protein Class Prediction Based on Transformer Encoders and Multi-Scale Convolutions

DeepEnzyme: a robust deep learning model for improved enzyme turnover number prediction by utilizing features of protein 3D-structures

Evidential deep learning for trustworthy prediction of enzyme commission number

Cross-Species Protein Function Prediction with Asynchronous-Random Walk

Enzyme function prediction using contrastive learning

ifDEEPre: large protein language-based deep learning enables interpretable and fast predictions of enzyme commission numbers

DeepAdd: Protein function prediction from k-mer embedding and additional features

Exploration and Evaluation of Machine Learning-Based Models for Predicting Enzymatic Reactions

A general model for predicting enzyme functions based on enzymatic reactions

Protein Function Prediction With Functional and Topological Knowledge of Gene Ontology

Enzyme Activity Prediction of Sequence Variants on Novel Substrates using Improved Substrate Encodings and Convolutional Pooling

DEEPred: Automated Protein Function Prediction with Multi-task Feed-forward Deep Neural Networks

A Novel Prediction Method for ATP-Binding Sites from Protein Primary Sequences Based on Fusion of Deep Convolutional Neural Network and Ensemble Learning.

ReactZyme: A Benchmark for Enzyme-Reaction Prediction