A hybrid machine learning model with attention mechanism and multidimensional multivariate feature coding for...

Area of Science:

• Bioinformatics
• Genomics
• Machine Learning

Background:

• Essential genes are critical for species survival and development.
• Identifying essential genes aids in understanding fundamental life processes and discovering disease therapeutic targets.
• Machine learning is a key approach for essential gene prediction, but faces challenges like feature extraction, data imbalance, and cross-species generalization.

Purpose of the Study:

• To develop a novel hybrid machine learning model for accurate essential gene prediction.
• To address challenges in feature extraction, data imbalance, and cross-species generalization in essential gene identification.
• To evaluate the model's performance and generalization capabilities across different species.

Main Methods:

• Proposed a hybrid model integrating Graph Convolutional Neural Networks (GCN) and Bi-directional Long Short-Term Memory (Bi-LSTM) with an attention mechanism.
• Employed multidimensional multivariate feature coding for gene sequences.
• Utilized GCN for feature encoding from visualized gene sequences and attention-based Bi-LSTM for feature importance assessment.

Main Results:

• The proposed EGP Hybrid-ML model achieved a sensitivity of 0.9122 in essential gene prediction.
• The model demonstrated superior predictive performance compared to existing methods.
• Cross-validation confirmed strong cross-species generalization capabilities.

Conclusions:

• The EGP Hybrid-ML model offers enhanced predictive accuracy and generalization for essential gene identification.
• This model has significant potential applications in bioinformatics, chemical information, and pharmaceutical research.
• Model code, architecture, parameters, and datasets are publicly available on GitHub.