ImmFinder: A Multiomics-Based Neural Network Approach for Predicting the Immune Genes in Livestock.

Area of Science:

• Genomics
• Immunology
• Bioinformatics

Background:

• Classifying immune and nonimmune genes in cattle is vital for disease resistance research.
• Traditional gene classification methods are time-consuming and labor-intensive.

Purpose of the Study:

• To introduce ImmFinder, a novel multimodal fully connected neural network (FCNN) framework for classifying immune genes in cattle.
• To leverage integrated genomic and transcriptomic datasets for improved gene classification.

Main Methods:

• Developed ImmFinder, a multimodal FCNN framework.
• Integrated genomic and transcriptomic datasets for training and evaluation.
• Utilized Python for implementation, with all scripts available on GitHub.

Main Results:

• ImmFinder achieved high accuracy (85.67%), F1-score (0.85), precision (0.86), and recall (0.85).
• AUC-ROC scores of 0.9250 (test) and 0.9264 (validation) demonstrate robust performance.
• The study highlights the effectiveness of multimodal deep learning for functional gene annotation.

Conclusions:

• ImmFinder offers a scalable approach for functional gene annotation by fusing multimodal data.
• Deep learning presents a powerful alternative to traditional methods in cattle functional genomics.
• Future work includes expanding datasets, data modalities, and experimental validation.