NIFtHool

NIFtHool identifies nitrogenase enzyme (NifH) proteins in amino acid sequences to detect diazotrophic microorganisms involved in atmospheric nitrogen fixation.

Key Features:

• Dataset: Training and evaluation used 4,911 NifH and 4,782 non-NifH protein sequences sourced from UniProt.
• Feature extraction: Implements k-mer counting and amino-acid embedding layers with a trainable convolutional layer that processes numerical vectors from amino acid chains to generate feature maps.
• Model: A deep neural network classifies amino acid sequences as NifH or non-NifH.
• Performance: Reported evaluation metrics are accuracy 96.4%, sensitivity 95.2%, and specificity 96.7%.

Scientific Applications:

• Microbial ecology: Enables identification of NifH proteins to study diazotrophic community composition and function.
• Agricultural biotechnology: Supports detection of nitrogen-fixing microorganisms relevant to soil fertility and crop productivity.
• Environmental science: Facilitates tracking of atmospheric nitrogen fixation activity via NifH protein detection in environmental samples.

Methodology:

Uses a UniProt-derived dataset (4,911 NifH, 4,782 non-NifH); extracts features via k-mer counting and embedding layers with a trainable convolutional layer that generates feature maps from numerical amino acid vectors; classifies sequences with a deep neural network and reports accuracy, sensitivity, and specificity.