Skip to main content

Alzheimer’s disease: using gene/protein network machine learning for molecule discovery in olive oil


Alzheimer’s disease (AD) poses a profound human, social, and economic burden. Previous studies suggest that extra virgin olive oil (EVOO) may be helpful in preventing cognitive decline. Here, we present a network machine learning method for identifying bioactive phytochemicals in EVOO with the highest potential to impact the protein network linked to the development and progression of the AD. A balanced classification accuracy of 70.3 ± 2.6% was achieved in fivefold cross-validation settings for predicting late-stage experimental drugs targeting AD from other clinically approved drugs. The calibrated machine learning algorithm was then used to predict the likelihood of existing drugs and known EVOO phytochemicals to be similar in action to the drugs impacting AD protein networks. These analyses identified the following ten EVOO phytochemicals with the highest likelihood of being active against AD: quercetin, genistein, luteolin, palmitoleate, stearic acid, apigenin, epicatechin, kaempferol, squalene, and daidzein (in the order from the highest to the lowest likelihood). This in silico study presents a framework that brings together artificial intelligence, analytical chemistry, and omics studies to identify unique therapeutic agents. It provides new insights into how EVOO constituents may help treat or prevent AD and potentially provide a basis for consideration in future clinical studies.


Alzheimer’s disease (AD) is a chronic neurodegenerative disorder that manifests clinically with a progressive decline in cognitive functions and memory. The disease is characterized primarily by the pathological presence of deposits of misfolded proteins (i.e., amyloid beta peptides and phosphorylated microtubule-associated tau protein) and cell loss in brain areas important for language and memory. Despite decades of research, the exact pathogenesis of AD remains unknown and there exist no treatments to reverse or prevent its neurocognitive decline [1]. For this reason, worldwide the disease exerts a profound impact on the affected individuals and the caregivers, society, and healthcare systems that support them. Given the profound human, social, and economic burden that AD causes, the limited understanding of its pathophysiology, and lack of effective treatment modalities, it behooves researchers to consider novel and unconventional therapies for prevention and treatment of this disease [2].

In recent years, there has been growing interest in the potential neuroprotective effects of extra virgin olive oil (EVOO). This stems from EVOO being a key component of the Mediterranean diet, the consumption of which has been associated with a lower incidence of dementia and cognitive decline in large epidemiological studies [3,4,5]. Moreover, several interventional clinical studies have also shown promising results in the use of olive oil and EVOO phytochemicals to improve cognitive outcomes in individuals with AD [6,7,8]. However, the specific phytochemicals responsible for these effects remain unknown due to the complex nature of EVOO [5, 7, 9], and the time and cost associated with isolating sufficient quantities of individual constituents and examining their effects on the development and progression of AD.

Different methods have been used to identify bioactive molecules in food that may have therapeutic potential, particularly for complex diseases like AD. One of the most prominent computational approaches involves investigation of bioactive food molecules based on their structural similarity to the approved or experimentally validated drugs for treatment. This approach assumes that similarities in the chemical structures of food molecules and drug molecules would result in similar biological effects [10]. However, this method has its limitations. Even minor changes in a molecule’s chemical structure can lead to significant differences in biological effects. Consequently, a subtly modified molecule may not possess the same therapeutic potential as the original compound, or it may even cause undesirable side effects. The other approach involves exploration of bioactive food molecules by examining their individual protein targets, an approach often referred to as “one disease–one target–one drug” [11]. This reductionist approach to therapeutic intervention development links a specific disease to a single target molecule (such as a protein or enzyme), which is then modulated by a single drug or food molecule to treat the disease. The underlying idea is that a disease can be understood as a consequence of a single molecular dysfunction, and that correcting this dysfunction can lead to the cure or management of the disease. In such a scenario, the drug or food molecule would interact specifically with the target molecule in a specific manner (e.g., inhibiting its activity, enhancing its activity, or modulating it in some other way) that intervenes with the disease process. This approach has proven to be successful in some cases, such as the development of targeted therapies for certain types of cancer. For example, pembrolizumab (Keytruda®) and nivolumab (Opdivo®) drugs target the programmed cell death protein 1 (PD-1) pathway, which helps cancer cells evade the immune system. They are used to treat various types of cancer, including melanoma, non-small cell lung cancer, and head and neck cancer [12]. Generally, “one disease–one target–one drug/food molecule” approach may be of more limited utility for complex and multifactorial diseases (i.e., involving multiple molecular pathways and genetic factors).

Recent advances in artificial intelligence (AI), coupled with the explosive growth of large-scale, multi-source data on food, drugs, and diseases, offer a unique opportunity to identify molecules within foods that may prevent and/or reverse disease by using more holistic and systems-based approaches [13]. Among these, network machine learning [14, 15] and graph neural networks [16] have shown promising results in predicting bioactive molecules within foods based on their ability to target disease network (i.e., dysregulated genes and protein pathways). An example of this approach is the virus–host interaction network responsible for COVID-19 [17].

In the present study, we hypothesize that a successful prevention or treatment strategy for AD should focus on the modulation of multiple biochemical networks involved in its pathogenesis including amyloid beta production and aggregation, tau hyperphosphorylation, and neuroinflammation [18]. Drawing on our prior research investigating bioactive compounds in food against cancer [16] and COVID-19 [17], we utilized network-based machine learning methods and human protein–protein interaction data to identify EVOO-based bioactive molecules that could target AD.


Our approach is based on the modeling of the interplay between disease-causing proteins and proteins disrupted by drugs or EVOO phytochemicals, while accounting for network interactions. In simple terms, human biological processes can be seen as a complex network of interacting genes and proteins, with humans having over 20 thousand protein-encoding genes and millions of protein–protein interactions. This complex network is known as an interactome [19].

In a healthy state, the system can be represented as a baseline or ground level. Disease-causing protein dysfunctions would disturb or disrupt their biochemical and signaling processes; this can be represented as a perturbation from the baseline. Such a disturbance would impact other proteins through protein–protein interactions, leading to a chain reaction, known as “network propagation” [14]. As a result, a single protein dysfunction can influence many other proteins and pathways in a cascade of perturbations. Different proteins will be impacted differently, depending on how “far” they are in the interaction network from the dysregulated protein. We have modeled the extent of these perturbations for over 20,000 proteins by using the known dysfunctional proteins linked to AD as starting perturbation points and employing a network propagation technique called “random walk with restarts” [20]. The resulting > 20,000 levels of protein disruption are referred to as a disease perturbation profile of the AD. Similarly, drugs or phytochemicals can interact with specific proteins and their effect on the ground state of those proteins can propagate through the network. In this case, the overall levels of perturbation of proteins are referred to as a drug or phytochemical perturbation profile (see Fig. 1A).

Fig. 1
figure 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

The methodology employed in the present study assumes that drugs and EVOO phytochemicals with anti-AD properties influence the proteins and pathways involved in the AD onset and progression. However, the disease and the drugs/phytochemicals do not have to target the exact same genes or protein, but rather share common network regions connected to them. To exert an anti-AD effect, a drug perturbation profile or phytochemical perturbation profile would be expected to be similar (but not necessarily identical) to the AD disease perturbation profile. A drug that does not affect AD (i.e., non-anti-AD) would be expected to have a perturbation profile that shares little overlap with the AD disease perturbation profile. To determine the level of similarity necessary to identify a unique drug/phytochemical candidate with anti-AD properties, the machine learning predictive model is trained using drugs expected to intervene with AD (anti-AD drugs, e.g., midostaurin) and drugs that do not affect AD (non-anti-AD drugs, e.g., enoximone) as positive class and negative class references, respectively (Fig. 1A, B and Sect. “Model training and application”).

AD protein-encoding gene selection

In the present study, we identified proteins involved in AD using the extensive public dataset called DisGeNET [21]. From this dataset, 3,397 protein-encoding genes were found and then manually refined using expert-curated databases to ensure high-quality data. The databases used included UniProt [22], Comparative Toxicogenomics Database [23], Orphanet [24], Clinical Genome Resource [25], Genomics England PanelApp [26], Cancer Genome Interpreter [27], and the Psychiatric Disorders Gene Association Network [28].

Our study concentrated on proteins that influence AD onset or development, excluding those marked as therapeutic to exclude circularity in the analysis. This selection process led to the identification of 101 “key” protein-encoding genes strongly associated with AD risk and progression. These “key” genes were then mapped onto the STRING database [16] which provides a list of known protein–protein interactions for various organisms, along with their confidence scores (0–999). This step resulted in identification of 73 “key” AD proteins (Additional File 1). Protein–protein interactions were filtered later by thresholding the confidences scores, as part of the machine learning process (described below in the Methods Sect. “Network propagation”).

EVOO phytochemical selection

We have developed a comprehensive list of 782 phytochemicals present in extra virgin olive products, including EVOO, table olives, and olive leaves (Additional File 2). This list includes plant-based chemicals and excludes metabolic intermediates, such as glucose [29]. Of these 782 phytochemicals, only those unique to EVOO or found in high concentrations were selected [30]. This selection was done by searching the literature for evidence of these chemicals in the olive oil matrix, a screening step that resulted in the identification of 117 EVOO “marker” phytochemicals.

DrugBank [31] and DrugCentral [32] databases were used to compile a list of experimental and available FDA-approved drugs, and FoodDB [33] database was used to compile a list of known phytochemicals (as described previously [16]). The STITCH database [34] was then used to find human proteins with which the molecules could interact, resulting in 67 EVOO marker phytochemicals. The database provides a confidence score (0–999) for each chemical–protein interaction, which is a measure of the likelihood of a true interaction based on the available evidence. In the present study, a fixed confidence cutoff of 200 was used, meaning that only interactions with a score of 200 or higher were considered. This threshold was chosen to ensure that the interactions being studied were more likely to be biologically relevant and supported by experimental evidence rather than only predicted computationally.

Network propagation

After compiling the lists of drugs and EVOO phytochemicals and their impacted genes/proteins, and identifying the dysfunctional proteins associated with AD, we employed our previously developed network propagator tool (used in cancer [16] and COVID-19 [17] projects) to perform “random walk with restarts” [20]. This step enabled us to generate drug, EVOO phytochemical and disease perturbation profiles based on the STRING protein–protein network, consisting of 1,048,574 connections between 20,256 human proteins (matched between STITCH and STRING databases).

The random walk propagation was executed independently for drugs/EVOO phytochemicals and for the AD interactomes, using a range of restart probability values and several protein–protein connection thresholds (STRING protein connection thresholds of 400, 600, and 800; restart probability parameter c ranging from 0.0001 to 1.0). This range was selected based on our previous experience with random walk propagation in application to cancer and COVID-19 [16, 17]. Pearson correlation coefficients were calculated to assess the extent of overlap between the AD perturbation profile and each of the drug or EVOO phytochemical perturbation profiles. For each molecule–disease pair, a total of 2,160 Pearson correlation coefficients were calculated with different propagation settings, allowing for the assessment of molecule–disease perturbation similarities.

Model training and application

Leveraging molecule–disease perturbation similarities, the machine learning models were trained to predict which FDA-approved drugs, of all those available in in DrugBank and DrugCentral (N = 1,802), are undergoing advanced stages (3 and 4) of FDA clinical trials (N = 32) for the treatment of AD. These 32 drugs are FDA-approved for other disease states and are being re-evaluated as potential therapeutic agents for the treatment of AD. A logistic regression classifier (based on the scikit-learn Python library for machine learning [35]) was fitted to the calculated Pearson correlation coefficients for each propagation settings from the previous step. The best settings were selected using fivefold stratified cross-validation. This approach is similar to the one we employed to predict phytochemicals with anticancer properties [16]. The overall predictive capacity of this approach was validated using 25 repeats of fivefold stratified cross-validation, with the optimal predictive model/propagation setting being selected by the internal fivefold stratified cross-validation.

A final set of models were identified that exhibited individual balanced accuracies higher than the cross-validation one; this resulted in 64 models. These models were then used to predict the probabilities of EVOO phytochemicals being able to exhibit anti-AD properties based on the correlation between their perturbation profiles and the perturbation profile of AD. Predicted probabilities were averaged between the selected models to provide a more robust ensemble-based probability prediction, further referred to as the correlation probability (%).

Pathway analysis

The methodology was completed using GSEA Prerank [36] to analyze the top 10 approved drugs and EVOO (extra virgin olive oil) phytochemicals predicted by our model to have the highest correlation probability. We examined significant KEGG 2023 pathways [18] (p value < 0.001) for each molecule and ranked them based on their p values. KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways are a collection of databases and tools that help researchers understand the biological functions, interactions, and networks within cells at a systems level. These pathways represent the molecular interactions and reaction networks for various cellular processes, such as metabolism, genetic information processing, environmental information processing, cellular processes, organismal systems, and human diseases. KEGG pathways help to visualize and interpret complex biological data, making it easier for researchers to understand how genes and proteins interact in different cellular contexts [18].

In this type of analysis, the pathway’s enrichment score measures the extent to which a particular biological process is overrepresented among the genes or proteins being studied. The nominal p value indicates the likelihood of observing the enrichment score by chance (assuming the gene/protein ranking is random and unrelated to the biological process under investigation) [18]. We calculated the interaction between AD proteins and EVOO phytochemical targets by multiplying the AD and EVOO phytochemical vectors. This product was then used in the GSEA analysis.

All data and Python scripts for the modeling, analysis, and figures are available from this Bitbucket repository:


Our model allowed us to predict with 70.3% ± 2.6% accuracy which previously FDA-approved drugs were in phase 3 and 4 trials for AD specifically versus all other FDA-approved drugs not in AD trials. The resulting 64 models were used for scoring the EVOO phytochemicals; the probabilities of these phytochemicals predicted to be similar to the drugs in FDA phase 3 and 4 trials were then averaged to produce the final consensus prediction. EVOO phytochemicals with the highest probability of being like compounds in FDA trials were considered most likely to be biologically active.

Using our model, the FDA-approved drugs for other diseases (i.e., non-anti-AD drugs) were re-evaluated and ranked according to their correlation probability as having the highest probability of affecting AD (Additional File 3). Analysis of the most common pathways targeted by these drugs was Alzheimer’s disease (10), Olfactory Transduction (10), Insulin Signaling (5), Phosphatidylinositol Signaling System (2) and Long-Term Potentiation (1). The numbers in parentheses represent the number of times each pathway was identified for these drugs with a statistically significant p value for involvement with AD.

Of the 67 EVOO marker phytochemicals (Additional File 4), quercetin, genistein, luteolin, palmitoleate, stearic acid, apigenin, epicatechin, kaempferol, squalene, and daidzein had the highest probability of interfering with AD (Table 1). The most common pathways predicted to be affected by these phytochemicals include Olfactory Transduction (10), Alzheimer’s disease (9), Insulin Signaling Pathway (7), Phosphatidylinositol Signaling System (5), and Vascular Smooth Muscle Contraction (3). As before, the numbers in parentheses represent the number of times each pathway was identified for these EVOO phytochemicals with a statistically significant p value for involvement with AD. Examples of the select EVOO phytochemicals and their connection with the most common AD pathways and involved proteins in our analysis are depicted in Fig. 2.

Table 1 EVOO marker phytochemicals predicted to have highest probability of influencing AD onset and/or progression
Fig. 2
figure 2

Predicted EVOO phytochemicals targeting dysregulated pathways in AD. The targeted pathways include: Blue denotes (Olfactory Transduction), violet denotes (Alzheimer’s disease), green denotes (Insulin Signaling), greenish blue denotes (Phosphatidylinositol Signaling), and orange denotes (Vascular Smooth Muscle Contraction). The key protein-encoding genes associated with a specific pathway (Additional File 5) are assigned the same color for easy identification. The size of each gene/protein node is proportional to how often the related pathway is targeted by the model EVOO phytochemicals listed in Table 1. Nodes associated with EVOO phytochemicals and their interactions with protein-encoding genes from AD dysregulated pathways are highlighted in brown. Only experimentally validated protein–protein interactions (STRING score greater than 800) are emphasized


Using machine learning, the present study identified FDA-approved drugs and EVOO phytochemicals that are likely to interact with genes and proteins associated with AD pathophysiology. The model was initially trained to differentiate between drugs approved by the FDA in phases 3 or 4 to treat AD (= positive class) versus those drugs approved by the FDA to treat other diseases aside from AD (= negative class). This was the model used to predict which EVOO phytochemicals might affect the development and/or progression of AD.

The correlation probability scores obtained for model EVOO phytochemicals were similar to those obtained for the FDA-approved phase 3 or 4 drugs, suggesting that the phytochemicals interact with the same pathways as the FDA-approved drugs.

The use of machine learning to identify drug targets using existing databases is only in its infancy and the accuracy of the predictions rests in the quality of the databases being used (which are always being updated as new knowledge is generated). The results obtained in the present study using network propagation (random walk with restarts) support the validity of this approach in that several of the identified EVOO phytochemicals, such as quercetin, have been previously found to have a positive effect on preventing the development or progression of AD (see discussion below). For example, quercetin, the EVOO phytochemical shown to have the strongest correlation probability in the present study, has previously been evaluated as a neuroprotective agent in AD [38]. It is thought to be protective against AD by decreasing oxidative stress, modulation of cytokines, inhibiting amyloid beta aggregation, and decreasing tau phosphorylation [37,38,39]. Genistein, apigenin, catechin (also called epicatechin), kaempferol, similarly, have also been identified as potential therapeutic agents for AD [40,41,42, 44,45,46,47,48,49,50]. The flavonoids luteolin and daidzein, however, are novel agents which have not been investigated as a therapeutic in AD. Palmitoleate and stearic acid have undergone limited evaluation for their biologic actions in the central nervous system [53,54,55,56]. To date, squalene has not been considered a therapeutic phytochemical, although its metabolism has been considered in the pathogenesis of AD [57,58,59].

Eleven different KEGG pathways were identified as being targeted by the ten EVOO marker phytochemicals deemed most likely to interact with AD (Table 1). The pathways with the highest number of occurrences included Olfactory Transduction, Alzheimer’s disease, Insulin Signaling Pathway, Phosphatidylinositol Signaling System, and Vascular Smooth Muscle Contraction, respectively. The Alzheimer’s disease pathway includes the accumulation of beta-amyloid peptides, the formation of neurofibrillary tangles, oxidative stress, inflammation, and neuronal apoptosis events [18]. Olfactory Transduction is not known to be directly related to AD, although there is some evidence to suggest that changes in olfactory function may be an early marker of the disease. For example, AD-affected individuals often have a reduced sense of smell [60, 61], and olfactory dysfunction may precede the onset of cognitive symptoms by several years [62]. In addition, there is evidence to suggest that the accumulation of amyloid beta protein (a hallmark of AD) may disrupt olfactory function by interfering with the olfactory transduction pathway [63]. The Insulin Signaling Pathway is involved in several aspects of AD pathology, including amyloid beta and tau metabolism, neuroinflammation, and synaptic plasticity [64]. Phosphatidylinositol Signaling System pathway plays a crucial role in the regulation of various cellular processes, such as cell growth, proliferation, and survival [65]. In AD, this system has been implicated in the regulation of amyloid beta production, neuroinflammation, and synaptic dysfunction, and, as such, its dysregulation may contribute to the pathological processes that underlie AD [65]. Finally, Vascular Smooth Muscle Contraction (VSMC) is crucial for blood flow and pressure regulation. It is implicated in AD progression, as cerebrovascular dysfunction can lead to impaired blood flow, reduced brain oxygen supply, and increased AD symptoms and neuronal damage [66]. Amyloid beta accumulation can also disrupt blood vessels and impair VSMC, limiting their response to blood flow and pressure changes and escalating the risk of hypoperfusion and neurovascular damage [67].

The analytical approach taken in the present study has several limitations, many of which are due to the nature of the datasets underlying our data inputs. For example, the exact pathogenesis of AD is unknown. If there are proteins and genes that have a profound impact on AD development but are as yet undiscovered, they will not be accounted for in this study. The STITCH and STRING databases are themselves incomplete, and while it is unlikely, it is possible there are proteins, phytochemical–protein, or protein–protein interactions known to have an impact in AD which are not included in those databases. As these databases improve with advancing knowledge, that approach taken in the present study should become more powerful. Furthermore, we only analyzed a subset of AD expert-curated genes; it is possible that the 73 we selected do not fully characterize the disease process. Our analysis also does not discern whether the identified EVOO phytochemicals are necessarily effective against AD or whether their effects would be anti-AD or pro-AD, although they are more likely to be as effective as the phase 3 and 4 FDA-approved drugs currently undergoing AD clinical trials than any other molecules found in EVOO. It is also important to recognize that the number of these positive class drugs (i.e., FDA-approved drugs undergoing AD trials) was much smaller (32) than number of negative class drugs (i.e., FDA-approved non-anti-AD drugs) that were used for model training (1,745). Such an unbalanced dataset increases the likelihood of bias toward the negative class, leading to underestimation of EVOO phytochemicals that may influence AD. With identification of more anti-AD drugs and updated databases in the future, it should be possible to further improve the current model’s predictive capacity.

We hope our in silico work will inspire further studies to experimentally address the existing limitations, to validate our findings, and to fully evaluate the EVOO role in the prevention of AD.


It is well known that diet and lifestyle influence health. Machine learning is a novel, cost-effective way to evaluate the potential health benefits of individual EVOO phytochemicals. The present study provides an approach that brings together artificial intelligence, analytical chemistry, and omics studies to explore the interactions of phytochemicals with pathways involved in a disease states, information that can lead to the identification of novel therapeutic entities in a natural product (that contains a heterogeneous mixture of phytochemicals). The analyses identified several individual EVOO phytochemicals that have a high likelihood of interfering with AD, a few of which (e.g., quercetin, genistein) have shown promising effects on AD pathogenesis. Others (e.g., luteolin) are worthy of further in vitro and in vivo study. It is only through the conduct of such studies will the predictive utility of our machine learning approach be validated. While the results of the present study shed light on how EVOO may help treat or prevent AD, the same approach may be applied to identify EVOO phytochemicals (or other food constituents) that treat other diseases, such as hypertension or dyslipidemia.


  1. Perneczky R. Dementia treatment versus prevention. Dialogues Clin Neurosci. 2019;21:43–51.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Shah H, Albanese E, Duggan C, Rudan I, Langa KM, Carrillo MC, et al. Research priorities to reduce the global burden of dementia by 2025. Lancet Neurol. 2016;15:1285–94.

    Article  PubMed  Google Scholar 

  3. Scarmeas N, Stern Y, Tang M-X, Mayeux R, Luchsinger JA. Mediterranean diet and risk for Alzheimer’s disease. Ann Neurol. 2006;59:912–21.

    Article  PubMed  PubMed Central  Google Scholar 

  4. Román GC, Jackson RE, Reis J, Román AN, Toledo JB, Toledo E. Extra-virgin olive oil for potential prevention of Alzheimer disease. Rev Neurol (Paris). 2019;175:705–23.

    Article  PubMed  Google Scholar 

  5. Vassallo N, editor. Natural compounds as therapeutic agents for Amyloidogenic diseases. Cham: Springer International Publishing; 2015.

    Google Scholar 

  6. Martínez-Lapiscina EH, Clavero P, Toledo E, Estruch R, Salas-Salvadó J, San Julián B, et al. Mediterranean diet improves cognition: the PREDIMED-NAVARRA randomised trial. J Neurol Neurosurg Psychiatry. 2013;84:1318–25.

    Article  PubMed  Google Scholar 

  7. Kaddoumi A, Denney TS, Deshpande G, Robinson JL, Beyers RJ, Redden DT, et al. Extra-virgin olive oil enhances the blood-brain barrier function in mild cognitive impairment: a randomized controlled trial. Nutrients. 2022;14(23):5102.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  8. Valls-Pedret C, Sala-Vila A, Serra-Mir M, Corella D, de la Torre R, Martínez-González MÁ, et al. Mediterranean diet and age-related cognitive decline: a randomized clinical trial. JAMA Intern Med. 2015;175:1094–103.

    Article  PubMed  Google Scholar 

  9. Rigacci S. Olive oil phenols as promising multi-targeting agents against Alzheimer’s disease. Adv Exp Med Biol. 2015;863:1–20.

    Article  CAS  PubMed  Google Scholar 

  10. Bolognesi ML, Banzi R, Bartolini M, Cavalli A, Tarozzi A, Andrisano V, et al. Novel class of quinone-bearing polyamines as multi-target-directed ligands to combat Alzheimer’s disease. J Med Chem. 2007;50:4882–97.

    Article  CAS  PubMed  Google Scholar 

  11. Santos R, Ursu O, Gaulton A, Bento AP, Donadi RS, Bologa CG, et al. A comprehensive map of molecular drug targets. Nat Rev Drug Discov. 2017;16:19–34.

    Article  CAS  PubMed  Google Scholar 

  12. Gong J, Chehrazi-Raffle A, Reddi S, Salgia R. Development of PD-1 and PD-L1 inhibitors as a form of cancer immunotherapy: a comprehensive review of registration trials and future considerations. J Immunother Cancer. 2018;6:8.

    Article  PubMed  PubMed Central  Google Scholar 

  13. David L, Thakkar A, Mercado R, Engkvist O. Molecular representations in AI-driven drug discovery: a review and practical guide. J Cheminform. 2020;12:56.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  14. Cowen L, Ideker T, Raphael BJ, Sharan R. Network propagation: a universal amplifier of genetic associations. Nat Rev Genet. 2017;18:551–62.

    Article  CAS  PubMed  Google Scholar 

  15. Ruiz C, Zitnik M, Leskovec J. Identification of disease treatment mechanisms through the multiscale interactome. Nat Commun. 2021;12:1796.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  16. Veselkov K, Gonzalez G, Aljifri S, Galea D, Mirnezami R, Youssef J, et al. HyperFoods: machine intelligent mapping of cancer-beating molecules in foods. Sci Rep. 2019;9:9237.

    Article  PubMed  PubMed Central  Google Scholar 

  17. Laponogov I, Gonzalez G, Shepherd M, Qureshi A, Veselkov D, Charkoftaki G, et al. Network machine learning maps phytochemically rich “Hyperfoods” to fight COVID-19. Hum Genomics. 2021;15:1.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  18. Kanehisa M, Furumichi M, Sato Y, Kawashima M, Ishiguro-Watanabe M. KEGG for taxonomy-based analysis of pathways and genomes. Nucleic Acids Res. 2023;51:D587–92.

    Article  CAS  PubMed  Google Scholar 

  19. Vidal M, Cusick ME, Barabási A-L. Interactome networks and human disease. Cell. 2011;144:986–98.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Valdeolivas A, Tichit L, Navarro C, Perrin S, Odelin G, Levy N, et al. Random walk with restart on multiplex and heterogeneous biological networks. Bioinformatics. 2019;35:497–505.

    Article  CAS  PubMed  Google Scholar 

  21. Piñero J, Saüch J, Sanz F, Furlong LI. The DisGeNET cytoscape app: exploring and visualizing disease genomics data. Comput Struct Biotechnol J. 2021;19:2960–7.

    Article  PubMed  PubMed Central  Google Scholar 

  22. Bateman A, Martin M-J, Orchard S, Magrane M, Agivetova R, Ahmad S, et al. UniProt: the universal protein knowledgebase in 2021. Nucleic Acids Res. 2021;49:D480–9.

    Article  Google Scholar 

  23. Davis AP, Grondin CJ, Johnson RJ, Sciaky D, Wiegers J, Wiegers TC, et al. Comparative toxicogenomics database (CTD): update 2021. Nucleic Acids Res. 2021;49:D1138–43.

    Article  CAS  PubMed  Google Scholar 

  24. Pavan S, Rommel K, Mateo Marquina ME, Höhn S, Lanneau V, Rath A. Clinical practice guidelines for rare diseases: the orphanet database. PLoS ONE. 2017;12:e0170365.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Rehm HL, Berg JS, Brooks LD, Bustamante CD, Evans JP, Landrum MJ, et al. ClinGen — the clinical genome resource. N Engl J Med. 2015;372:2235–42.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  26. Martin AR, Williams E, Foulger RE, Leigh S, Daugherty LC, Niblock O, et al. PanelApp crowdsources expert knowledge to establish consensus diagnostic gene panels. Nat Genet. 2019;51:1560–5.

    Article  CAS  PubMed  Google Scholar 

  27. Tamborero D, Rubio-Perez C, Deu-Pons J, Schroeder MP, Vivancos A, Rovira A, et al. Cancer genome Interpreter annotates the biological and clinical relevance of tumor alterations. Genome Med. 2018;10:25.

    Article  PubMed  PubMed Central  Google Scholar 

  28. Gutiérrez-Sacristán A, Grosdidier S, Valverde O, Torrens M, Bravo À, Piñero J, et al. PsyGeNET: a knowledge platform on psychiatric disorders and their genes. Bioinformatics. 2015;31:3075–7.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Kalogiouri NP, Aalizadeh R, Thomaidis NS. Application of an advanced and wide scope non-target screening workflow with LC-ESI-QTOF-MS and chemometrics for the classification of the greek olive oil varieties. Food Chem. 2018;256:53–61.

    Article  CAS  PubMed  Google Scholar 

  30. Kalogiouri NP, Aalizadeh R, Dasenaki ME, Thomaidis NS. Authentication of greek PDO kalamata table olives: a novel non-target high resolution mass spectrometric approach. Molecules. 2020;25:2919.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  31. Wishart DS, Feunang YD, Guo AC, Lo EJ, Marcu A, Grant JR, et al. DrugBank 5.0: a major update to the DrugBank database for 2018. Nucleic Acids Res. 2018;46:D1074–82.

    Article  CAS  PubMed  Google Scholar 

  32. Avram S, Wilson TB, Curpan R, Halip L, Borota A, Bora A, et al. DrugCentral 2023 extends human clinical data and integrates veterinary drugs. Nucleic Acids Res. 2023;51:D1276–87.

    Article  CAS  PubMed  Google Scholar 

  33. Harrington RA, Adhikari V, Rayner M, Scarborough P. Nutrient composition databases in the age of big data: foodDB, a comprehensive, real-time database infrastructure. BMJ Open. 2019;9:e026652.

    Article  PubMed  PubMed Central  Google Scholar 

  34. Szklarczyk D, Santos A, von Mering C, Jensen LJ, Bork P, Kuhn M. STITCH 5: augmenting protein-chemical interaction networks with tissue and affinity data. Nucleic Acids Res. 2016;44:D380–4.

    Article  CAS  PubMed  Google Scholar 

  35. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: machine learning in python. J Mach Learn Res. 2012;12:2825.

    Google Scholar 

  36. Fang Z, Liu X, Peltz G. GSEApy: a comprehensive package for performing gene set enrichment analysis in Python. Bioinformatics. 2023;39:757.

    Article  Google Scholar 

  37. Ansari MA, Abdul HM, Joshi G, Opii WO, Butterfield DA. Protective effect of quercetin in primary neurons against Abeta(1–42): relevance to Alzheimer’s disease. J Nutr Biochem. 2009;20:269–75.

    Article  CAS  PubMed  Google Scholar 

  38. Khan H, Ullah H, Aschner M, Cheang WS, Akkol EK. Neuroprotective effects of quercetin in Alzheimer’s disease. Biomolecules. 2019;10:59.

    Article  PubMed  PubMed Central  Google Scholar 

  39. Zaplatic E, Bule M, Shah SZA, Uddin MS, Niaz K. Molecular mechanisms underlying protective role of quercetin in attenuating Alzheimer’s disease. Life Sci. 2019;224:109–19.

    Article  CAS  PubMed  Google Scholar 

  40. Uddin MS, Kabir MT. Emerging signal regulating potential of genistein against Alzheimer’s disease: a promising molecule of interest. Front Cell Dev Biol. 2019;7:197.

    Article  PubMed  PubMed Central  Google Scholar 

  41. Devi KP, Shanmuganathan B, Manayi A, Nabavi SF, Nabavi SM. Molecular and therapeutic targets of genistein in Alzheimer’s disease. Mol Neurobiol. 2017;54:7028–41.

    Article  CAS  PubMed  Google Scholar 

  42. Bagheri M, Joghataei M-T, Mohseni S, Roghani M. Genistein ameliorates learning and memory deficits in amyloid β(1–40) rat model of Alzheimer’s disease. Neurobiol Learn Mem. 2011;95:270–6.

    Article  CAS  PubMed  Google Scholar 

  43. Frigolet ME, Gutiérrez-Aguilar R. The role of the novel lipokine palmitoleic acid in health and disease. Adv Nutr. 2017;8:173S-181S.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. Zhao L, Wang J-L, Liu R, Li X-X, Li J-F, Zhang L. Neuroprotective, anti-amyloidogenic and neurotrophic effects of Apigenin in an Alzheimer’s disease mouse model. Molecules. 2013;18:9949–65.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  45. Venigalla M, Sonego S, Gyengesi E, Münch G. Curcumin and Apigenin - novel and promising therapeutics against chronic neuroinflammation in Alzheimer′s disease. Neural Regen Res. 2015;10:1181.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Balez R, Steiner N, Engel M, Muñoz SS, Lum JS, Wu Y, et al. Neuroprotective effects of apigenin against inflammation, neuronal excitability and apoptosis in an induced pluripotent stem cell model of Alzheimer’s disease. Sci Rep. 2016;6:31450.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  47. George RC, Lew J, Graves DJ. Interaction of cinnamaldehyde and epicatechin with tau: implications of beneficial effects in modulating Alzheimer’s disease pathogenesis. J Alzheimers Dis. 2013;36:21–40.

    Article  CAS  PubMed  Google Scholar 

  48. Zeng Y-Q, Wang Y-J, Zhou X-F. Effects of (-)Epicatechin on the pathology of APP/PS1 transgenic mice. Front Neurol. 2014;5:69.

    Article  PubMed  PubMed Central  Google Scholar 

  49. Beg T, Jyoti S, Naz F, Ali F, Ali SK, Reyad AM, Siddique YH. Protective effect of Kaempferol on the transgenic drosophila model of Alzheimer’s disease. CNS Neurol Disord-Drug. 2018;17(6):421–9.

    Article  CAS  Google Scholar 

  50. Silva dos Santos J, Gonçalves Cirino JP, de Oliveira Carvalho P, Ortega MM. The pharmacological action of Kaempferol in Central nervous system diseases: a review. Front Pharmacol 2021;11: 565700.

  51. Pan M, Han H, Zhong C, Geng Q. Effects of genistein and daidzein on hippocampus neuronal cell proliferation and BDNF expression in H19–7 neural cell line. J Nutr Health Aging. 2012;16:389–94.

    Article  CAS  PubMed  Google Scholar 

  52. Choi RCY, Zhu JTT, Yung AWY, Lee PSC, Xu SL, Guo AJY, et al. Synergistic action of flavonoids, Baicalein, and Daidzein in estrogenic and neuroprotective effects: a development of potential health products and therapeutic drugs against Alzheimer’s disease. Evid Based Complement Altern Med. 2013;2013:1–10.

    Google Scholar 

  53. Snigdha S, Astarita G, Piomelli D, Cotman CW. Effects of diet and behavioral enrichment on free fatty acids in the aged canine brain. Neuroscience. 2012;202:326–33.

    Article  CAS  PubMed  Google Scholar 

  54. Hosseini M, Poljak A, Braidy N, Crawford J, Sachdev P. Blood fatty acids in Alzheimer’s disease and mild cognitive impairment: a meta-analysis and systematic review. Ageing Res Rev. 2020;60:101043.

    Article  CAS  PubMed  Google Scholar 

  55. Otoki Y, Kato S, Nakagawa K, Harvey DJ, Jin L-W, Dugger BN, et al. Lipidomic analysis of postmortem prefrontal cortex phospholipids reveals changes in choline plasmalogen containing Docosahexaenoic acid and stearic acid between cases with and without Alzheimer’s disease. Neuromolecular Med. 2021;23:161–75.

    Article  CAS  PubMed  Google Scholar 

  56. Fraser T, Tayler H, Love S. Fatty acid composition of frontal, temporal and parietal neocortex in the normal human brain and in Alzheimer’s disease. Neurochem Res. 2010;35:503–13.

    Article  CAS  PubMed  Google Scholar 

  57. Varma VR, Büşra Lüleci H, Oommen AM, Varma S, Blackshear CT, Griswold ME, et al. Abnormal brain cholesterol homeostasis in Alzheimer’s disease—a targeted metabolomic and transcriptomic study. NPJ Aging Mech Dis. 2021;7:11.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  58. Zarrouk A, Debbabi M, Bezine M, Karym EM, Badreddine A, Rouaud O, et al. Lipid biomarkers in Alzheimer’s disease. Curr Alzheimer Res. 2018;15:303–12.

    Article  CAS  PubMed  Google Scholar 

  59. Hottman DA, Li L. Protein prenylation and synaptic plasticity: implications for Alzheimer’s disease. Mol Neurobiol. 2014;50:177–85.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Devanand DP, Michaels-Marston KS, Liu X, Pelton GH, Padilla M, Marder K, et al. Olfactory deficits in patients with mild cognitive impairment predict Alzheimer’s disease at follow-up. Am J Psychiatry. 2000;157:1399–405.

    Article  CAS  PubMed  Google Scholar 

  61. Murphy C, Bacon AW, Bondi MW, Salmon DP. Apolipoprotein E status is associated with Odor identification deficits in nondemented older persons. Ann NY Acad Sci. 1998;855(1):744–50.

    Article  CAS  PubMed  Google Scholar 

  62. Devanand DP, Lee S, Manly J, Andrews H, Schupf N, Doty RL, et al. Olfactory deficits predict cognitive decline and Alzheimer dementia in an urban community. Neurology. 2015;84:182–9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  63. Wesson DW, Levy E, Nixon RA, Wilson DA. Olfactory dysfunction correlates with amyloid-β burden in an Alzheimer’s disease mouse model. J Neurosci. 2010;30:505–14.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  64. de la Monte SM, Wands JR. Alzheimer’s disease is type 3 diabetes—evidence reviewed. J Diabetes Sci Technol. 2008;2:1101–13.

    Article  PubMed  PubMed Central  Google Scholar 

  65. Berman DE, Dall’Armi C, Voronov SV, McIntire LB, Zhang H, Moore AZ, Staniszewski A, Arancio O, Kim TW, Di Paolo G. Oligomeric amyloid-β peptide disrupts phosphatidylinositol-4, 5-bisphosphate metabolism. Nat Neurosci. 2008;11(5):547–54.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  66. Iadecola C. Neurovascular regulation in the normal brain and in Alzheimer’s disease. Nat Rev Neurosci. 2004;5:347–60.

    Article  CAS  PubMed  Google Scholar 

  67. Sweeney MD, Sagare AP, Zlokovic BV. Blood–brain barrier breakdown in Alzheimer disease and other neurodegenerative disorders. Nat Rev Neurol. 2018;14:133–50.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references


The computational/AI research team (KV, IL, DV, LR, GG) has been funded by the Vodafone Foundation with the projects of CORONA-AI/DRUGS DreamLab, UK Research and Innovation (under grant number 10058099), ERC Proof of concept Hyperfoods grant (899932) and the European Union (under grant number 101095359). The authors were also supported by the ERC-Consolidator Grant No. 724228 (LEMAN).


Funding was provided by Fundação para a Ciência e a Tecnologia, 2021.05460.BD; ERC-Consolidator, 724228 (LEMAN); ERC Proof of Concept, 899932 (Hyperfoods); UK Research and Innovation, 10058099; European Union, 101095359; and Vodafone Foundation, CORONA-AI/DRUGS DreamLab.

Author information

Authors and Affiliations



KV and VV contributed to design of research. KV and IL were involved in computational/AI project supervision. RA, NT, IL, DV, GG, NN, and LR contributed to data collection. LR, IL, and NN performed the project. LR, NN, IL, and KV wrote the manuscript. All authors discussed the results and commented on the manuscript.

Corresponding authors

Correspondence to Vasilis Vasiliou or Kirill Veselkov.

Ethics declarations

Ethics approval and consent to participate

Drs Vasiliou, Veselkov, and Thomaidis are serving as Editors for the Journal of Human Genomics and will exclude themselves from the review process.

Competing interests

The authors declare no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1

List of 73 AD genes considered to be dysregulated in AD.

Additional file 2

List of 782 phytochemicals present in extra virgin olive products, including EVOO, table olives, and olive leaves. It includes both EVOO markers and non-markers.

Additional file 3

10 FDA-approved drugs and not undergoing AD clinical trials that were classified the highest by our model.

Additional file 4

67 EVOO marker phytochemicals predicted and ranked according to the highest probability of influencing AD onset and/or progression.

Additional file 5

Detailed examination of the pathways targeted by phytochemicals in EVOO, which are associated with AD and have been highly predicted by our network propagation model.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Rita, L., Neumann, N.R., Laponogov, I. et al. Alzheimer’s disease: using gene/protein network machine learning for molecule discovery in olive oil. Hum Genomics 17, 57 (2023).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: