EssTFNet: integration of adaptive time-frequency and DNA language models for interpretable human essential gene...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Essential genes are critical for organism survival, making their identification vital for understanding life's origins and identifying therapeutic targets.
• Predicting essential genes is crucial for basic research and biomedical applications.

Purpose of the Study:

• To develop a novel, interpretable deep learning framework, EssTFNet, for accurate prediction of human essential genes.
• To enable mechanistic biological interpretation of gene essentiality predictions.

Main Methods:

• EssTFNet combines adaptive time-frequency analysis with a DNA language model, adapting the ATFNet architecture.
• DNA and protein sequences are mapped to time-series signals to extract periodic and nonstationary features.
• Feature selection and architectural optimization were employed for accuracy, interpretability, and generalization.

Main Results:

• EssTFNet achieved superior performance on the S1 benchmark task, outperforming mainstream sequence-based deep learning methods.
• Achieved an area under the curve (AUC) of 0.9679 and an area under the precision-recall curve (AUPRC) of 0.8491.
• The DeepLIFT attribution method identified functional motifs linked to gene essentiality, providing insights for experimental validation.

Conclusions:

• EssTFNet presents a powerful and interpretable deep learning framework for predicting human essential genes.
• The study offers a valuable methodological approach for future research and applications in genomics and personalized medicine.
• A web server and source code are available, facilitating broader research use.