NifFinder: improved Nif protein prediction using SWeeP vectors and neural networks

Area of Science:

• Microbiology and Bioinformatics
• Molecular Biology and Genetics

Background:

• Biological nitrogen fixation is crucial for ecosystems and agriculture.
• Identifying nitrogen fixation (Nif) proteins is challenging due to gene diversity and complexity.
• Existing tools often miss many Nif protein classes.

Purpose of the Study:

• To develop a comprehensive computational framework for predicting a wider range of Nif proteins.
• To improve the accuracy and reliability of Nif protein identification in microbial genomes.

Main Methods:

• Developed NifFinder, integrating SWeeP vector encoding with neural network classifiers.
• Designed to predict up to 24 different Nif protein classes across Archaea and Bacteria.
• Utilized benchmarking against curated Nif resources to validate performance.

Main Results:

• NifFinder achieved an average accuracy of 84.31% across 24 Nif protein classes.
• Demonstrated high sensitivity (86.49%), precision (81.97%), and F1-score (82.33%).
• Showed robust classification performance, even with imbalanced datasets.

Conclusions:

• NifFinder offers a more comprehensive and reliable method for genome-wide Nif protein identification.
• Expands predictive scope beyond traditional subsets of nif genes.
• Supports evolutionary insights and applications in agricultural sustainability.