Towards reconstruction of the human interactome from positive and negative experimental evidence

Protein-protein interactions (PPIs) have been extensively studied, but the human PPI network remains incomplete and biased. The problem stems from a lack of high-confidence positive interactions and an absence of data on likely non-interacting protein pairs. Currently, this information is not reported in PPI studies. To address this, researchers have developed a methodology to reconstruct this missing knowledge.

The reconstruction process involves mapping the experimental search space in which PPI screens have been conducted. By doing so, the team creates a model that estimates the likelihood of a PPI being real based on its testing and observation frequency. The researchers argue that incorporating negative protein pairs can help estimate the error rates of both experimental and computational screens.

The team demonstrates how this newfound knowledge can be utilized for calibration purposes. They further evaluate a simple machine learning approach to PPI prediction and propose a strategy for using negative data in training instead of relying on random protein pairs. The results of their study indicate that reconstructing the experimental search space unveils a previously overlooked layer of information from existing PPI data. This information can aid in creating a more accurate and complete mapping of the human interactome.