Biomarker Identification via a Factorization Machine-Based Neural Network With Binary Pairwise Encoding.

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics
• Genomics

Background:

• MicroRNAs (miRNAs) and long non-coding RNAs (lncRNAs) are crucial biomolecules involved in fundamental biological processes.
• Dysregulation of these non-coding RNAs is linked to complex human diseases, highlighting their potential as disease biomarkers.
• Accurate identification of disease-related biomarkers is essential for effective diagnosis, treatment, prognosis, and prevention strategies.

Purpose of the Study:

• To develop and evaluate a novel computational model for identifying disease-related biomarkers.
• To leverage deep learning and factorization machines for enhanced biomarker identification.
• To address the challenge of feature interdependence in biomarker discovery.

Main Methods:

• A factorization machine-based deep neural network with binary pairwise encoding (DFMbpe) was proposed.
• Binary pairwise encoding was employed to capture feature interdependence, even for features not co-occurring in samples.
• The model integrates low-order feature interactions (factorization machine) and high-order interactions (deep neural network).

Main Results:

• DFMbpe demonstrated superior performance compared to state-of-the-art models in identifying disease-related biomarkers.
• The model achieved significant improvements on both cross-validation and independent dataset evaluations.
• Case studies further validated the model's effectiveness in practical biomarker identification scenarios.

Conclusions:

• The DFMbpe model offers a powerful and effective approach for disease-related biomarker identification.
• The integration of binary pairwise encoding and a hybrid deep learning architecture captures complex feature interactions.
• This advancement holds significant promise for improving the diagnosis, treatment, and understanding of human diseases through biomarker discovery.