An Improved Algorithm and Web Application for Predicting Co-Complexed Proteins from Affinity Purification – Mass Spectrometry Data

摘要

Protein-protein interactions defined by affinity purification and mass spectrometry (APMS) approaches suffer from high false discovery rates. Consequently, the candidate interaction lists must be pruned of contaminants before network construction and interpretation, historically an expensive and time-intensive task. In recent years, numerous computational methods have been developed to identify genuine interactions from hundreds revealed by APMS experiments. Here, comparative analysis of several popular algorithms revealed complementarity in their classification accuracies, which is supported by their divergent scoring strategies. As such, we used two accurate and computationally efficient methods as features for machine learning using the Random Forest algorithm. Additionally, we developed novel mathematical models to include a variety of indirect data, such as mRNA co-expression, gene ontologies and homologous protein interactions as features within the classification problem. We show that our method, which we call Spotlite, outperforms existing methods on four diverse and public APMS datasets. Because implementation of existing APMS scoring methods requires computational expertise beyond many laboratories, we created a user-friendly and fast web application for APMS data scoring, analysis, annotation and network visualization, for use on new and existing data (). The utility of Spotlite and its visualization platform for revealing physical, functional and disease-relevant characteristics within APMS data is established through a focused analysis of the KEAP1 E3 ubiquitin ligase.