Fig. 1

Summary of the methodology used to identify EVOO phytochemicals that potentially disrupt the interactome associated with AD onset and progression. Panel A illustrates the core principles of network propagation applied to generate protein perturbation profiles for disease, drugs, and EVOO phytochemicals. Human interactome is shown as a graph of interconnected nodes (numbered circles), where each node represents a protein and the edges (connections) represent protein–protein interactions. Numbers are arbitrarily chosen to serve as unique protein identifiers to allow for quick visual cross-referencing. Proteins directly affected by drugs or EVOO compounds were identified from the STITCH database and proteins directly dysregulated in AD were identified using DisGeNET database (as indicated by arrows on the left-hand side of panel A). The extent of a protein’s perturbation is reflected by the intensity of the color associated with the protein circle for AD (blue), an anti-AD drug in late-stage clinical trials (i.e., a positive class drug example like midostaurin) (yellow), a non-anti-AD drug (i.e., a negative class drug example like enoximone) (cyan), and the query EVOO compound (like quercetin) whose probability of being effective as anti-AD drug we are aiming to predict (magenta). Network propagation section illustrates how perturbation propagates through protein–protein connections in the network. The further the propagation spreads through the network, the less perturbation is generally expected unless many paths lead to the same hub proteins, amplifying the perturbation of these proteins. Color intensity indicates the level of perturbation. Highly correlated profile regions are shown in red ovals and the poorly correlated profile regions are shown in blue ovals. Panel B depicts an overall machine learning-based approach for predicting EVOO phytochemicals targeting the AD interactome in a similar way as do the advanced-stage anti-AD drugs in clinical trials. A logistic regression classifier was trained to discriminate between anti-AD drugs (positive class) and non-anti-AD drugs (negative class) based on profile correlations, which was then used to predict the probability of EVOO phytochemicals exhibiting anti-AD properties