Optimizing Protein-Protein Interaction Filtering Based on Detection Methods to Construct High-Quality Interactomes for Network Medicine

<p xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" class="first" dir="auto" id="d808307e101">The protein-protein interaction (PPI) network, or interactome, is the foundation of Network Medicine, capturing complex molecular interactions within cells. A high-quality interactome is essential for identifying disease modules -clusters of proteins affected by pathogenic mutations-and predicting novel treatments. However, constructing a reliable interactome remains challenging due to the varying quality across detection methods: low-throughput approaches (X-ray crystallography, cryo-EM) yield high-confidence interactions but limited coverage, while high-throughput methods (tandem affinity purification, yeast two-hybrid) offer broader coverage but more false positives. <p xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" dir="auto" id="d808307e103">This study evaluated multiple filtering strategies for optimizing PPI quality based on the PPI detection method. Using data aggregated from IntAct, BioGRID, and IID via the NeDRexframework, we compared three approaches: (i) assigning confidence scores to detection methods, (ii) filtering based on the number of different methods identifying each interaction, and (iii) combining both reliability scores and detection frequency. We benchmarked the resulting networks against four state-of-the-art reference interactomes. <p xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" dir="auto" id="d808307e105">Our results show that assigning curated quality scores to detection methods and summing these scores for each interaction yields the most effective filtering strategy, balancing both confidence and coverage. Our findings provide a systematic framework for integrating heterogeneous PPI data into high-quality interactomes essential for advancing Network Medicine applications.