ThermoFinder: A sequence-based thermophilic proteins prediction framework.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Thermophilic proteins are crucial for research and industry.
• Existing computational methods for identifying thermophilic proteins have limitations due to data quality and model efficiency.
• There is a need for improved predictive models for thermophilic proteins.

Purpose of the Study:

• To develop a novel, sequence-based computational framework for predicting thermophilic proteins.
• To evaluate the performance of the proposed framework against existing state-of-the-art tools.
• To enable the regression-based prediction of temperature optimum values directly from protein sequences.

Main Methods:

• Development of a novel sequence-based prediction framework named ThermoFinder.
• Utilizing benchmark and newly constructed datasets for model training and evaluation.
• Employing feature ablation experiments to confirm the effectiveness of the approach.
• Applying Shapley Additive Explanations (SHAP) for feature importance analysis.

Main Results:

• ThermoFinder significantly outperforms previous state-of-the-art tools on benchmark datasets.
• Feature ablation experiments validate the contribution of different components of the ThermoFinder approach.
• ThermoFinder demonstrates high performance and consistency on newly constructed datasets, including one for regression-based temperature optimum prediction.
• SHAP analysis confirms the advantages of the features utilized by ThermoFinder.

Conclusions:

• ThermoFinder is a highly effective and comprehensive framework for predicting thermophilic proteins.
• The model's open-source availability facilitates its adoption in academic research and industrial applications.
• This work addresses the limitations of previous methods by providing an efficient and accurate prediction tool.