Data Imputation in Epistatic MAPs by Network-Guided Matrix Completion.

Area of Science:

• Genetics
• Computational Biology
• Bioinformatics

Background:

• Epistatic miniarray profile (E-MAP) is a key platform for discovering genetic interactions.
• E-MAPs provide quantitative data for precise detection of subtle interactions.
• Current E-MAP studies have missing measurements for up to 40% of gene pairs due to technological limitations.

Purpose of the Study:

• To introduce a novel computational method, network-guided matrix completion (NG-MC), for imputing missing genetic interaction data.
• To enhance the completeness of genetic interaction profiles for downstream analyses.
• To enable prediction of interactions for genes not originally included in E-MAP assays.

Main Methods:

• NG-MC utilizes low-rank probabilistic matrix completion integrated with prior knowledge from gene networks.
• The method assumes transitive interactions, inferring latent profiles based on gene network neighbors.
• Latent profiles are propagated through networks, updating gene network weights for improved prediction accuracy.

Main Results:

• NG-MC significantly outperformed existing imputation techniques across four diverse E-MAP datasets.
• The method effectively incorporated information from protein-protein interaction and gene ontology similarity networks.
• NG-MC successfully predicted interactions for genes absent from the original E-MAP experiments.

Conclusions:

• NG-MC offers a robust solution for recovering missing genetic interaction data in E-MAP studies.
• Integrating network information enhances the accuracy and scope of genetic interaction profiling.
• This approach expands the utility of E-MAP data by enabling predictions beyond the initially assayed gene pairs.