Denoising large-scale biological data using network filters.

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• Large-scale biological datasets are susceptible to noise from biological and technical sources.
• Noise can hinder accurate inference of underlying biological processes.
• Accurate data analysis is crucial for advancing biological understanding.

Purpose of the Study:

• To develop a general method for automatically reducing noise in large-scale biological data.
• To improve the recovery of underlying biological signals from noisy data.
• To enhance the accuracy of machine learning applications in biology.

Main Methods:

• Utilizes interaction networks to identify correlated/anti-correlated measurements.
• Applies network filters to combine or "filter" measurements, analogous to image denoising.
• Explores applying a single filter to the entire system or decomposing the system into modules with distinct filters.

Main Results:

• Network filters accurately reduce noise across various noise levels and structures in synthetic data.
• Network filtering increased accuracy by up to 43% in predicting human protein expression changes between healthy and cancerous tissues.
• The method effectively accounts for both correlation and anti-correlation between measurements.

Conclusions:

• Network filters offer a generalizable approach for denoising biological data.
• Partitioning networks before filtering significantly reduces errors in heterogeneous data, outperforming existing diffusion-based methods.
• Network filters demonstrate broad potential utility in systems biology applications, particularly with proteomics data.