Related Experiment Video
Updated: May 16, 2025
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
918
Strategies to include prior knowledge in omics analysis with deep neural networks
Kisan Thapa1, Meric Kinali1, Shichao Pei1
1Computer Science Department, University of Massachusetts Boston, 100 Morrissey Boulevard, Boston, MA 02125, USA.
Patterns (New York, N.Y.)
|April 4, 2025
Summary
Integrating biological prior knowledge into deep learning models enhances phenotype prediction from high-dimensional molecular profiles. This approach combats overfitting and improves model generalizability and interpretability in molecular biology research.
Area of Science:
- Molecular biology
- Bioinformatics
- Machine learning
Background:
- High-throughput molecular profiling generates vast datasets for biological research.
- Machine learning, particularly deep learning, is used for phenotype prediction from molecular data.
- High dimensionality and small sample sizes in molecular data lead to overfitting in deep learning models.
Purpose of the Study:
- To describe strategies for incorporating biological prior knowledge into deep learning models for molecular profile analysis.
- To review deep learning architectures, including graph neural networks, for this task.
Main Methods:
- Review of three major strategies for using prior biological knowledge in deep learning.
- Discussion of relevant deep learning architectures, including graph neural networks.
Main Results:
- Prior knowledge integration regularizes deep learning models.
- Incorporating prior knowledge improves generalizability and interpretability of predictions.
- Deep learning models, especially graph neural networks, show promise for molecular data analysis.
Conclusions:
- Biological prior knowledge is crucial for effective deep learning on molecular profiles.
- Strategies exist to integrate this knowledge, enhancing predictive model performance.
- Further research into architectures like graph neural networks is warranted.
