The answer might be yes: Machine-learning-based techniques identify bioactive molecules in foods with antiviral properties
Persistent outbreaks of COVID-19 are having damaging effects on global economies, healthcare, education, mental health and society as a whole. Tremendous efforts have been made in the development of vaccines. The approval of several vaccines in the second half of 2020 represented the beginning of the end of the pandemic. However, the strain on the healthcare system and individuals’ health remains until vaccination rollout has been completed.
There is a clear course of treatment for those severely affected, which involves hospitalization and prescription of recently approved or repurposed drugs against COVID-19. However, with a hospitalization ratio of only 3.5% , thousands of people contracting COVID-19 are not sufficiently ill to require hospitalization but find themselves struggling to manage their symptoms, as there is not a clear course of action apart from resting and taking over-the-counter medicines to relieve flu-like symptoms.
However, clues of a path for those struggling with COVID-19 in solitary might be found in foods or, to be more precise, in hyperfoods. In a paper published earlier this year, we applied graph machine learning to predict molecules within foods with anti-COVID-19 properties based on their capacity to target the SARS-CoV-2-host gene-gene (protein-protein) interactome. To this aim, we first calibrated a machine learning algorithm to predict anti-COVID-19 candidates among experimental and clinically approved drugs.
Clues of a path for those struggling with COVID-19 in solitary might be found in foods or, to be more precise, in hyperfoods
To calibrate the machine learning model, drug-protein and SARS-CoV-2-host protein interactions were represented as binary signals on the human protein-protein interaction (PPI) network and a network diffusion algorithm was applied to model the systemic genome-wide response to the drug and disease interventions. The Pearson correlation coefficient between the diffused profiles of drug compounds and COVID-19 disease was used to rank compounds targeting SARS-CoV-2-host interactome networks.
Drug-protein and SARS-CoV-2-host protein interactions on the PPI (left) were propagated using a diffusion algorithm to model the genome-wide response to the drug and disease intervention (right)
Parameters of the machine learning algorithm were calibrated for predicting experimentally validated drugs against COVID-19, task in which the model achieved an accuracy of 80–84.9% in a cross-validation setting. After model calibration, almost 8000 bioactive molecules within foods were run through the calibrated machine learning algorithm, which identified 52 biologically active molecules predicted to target SARS-CoV-2-host interactome networks.
52 biologically active molecules predicted to target SARS-CoV-2-host interactome networks
Prediction of “dark matter” of food biochemistry with anti-COVID-19 properties
Nutritional agencies recognize the value of plant-based foods, however, they typically focus on tracking minerals, vitamins and macro-nutrients, overlooking non-nutritional molecules within foods that can potentially exert protective or disease-beating effects. Therefore, these compounds are often regarded as the “dark matter” of nutritional science. However, they are molecules of paramount importance.
Anti-COVID-19 molecules identified belong to a variety of chemical classes, belonging to this “dark matter”, including (iso)flavonoids, terpenoids, phenols and indoles. It’s interesting to note that, due to their bitter taste, the food industry routinely removes some of these compounds through selective breeding or processes to improve taste.
Image source: Laponogov et al. — link
Hyperfoods to fight COVID-19
A key limitation on current food recommendations is the focus on individual phytochemicals within foods, where certain foods are marketed as “superfoods” due to their high content on a given phytochemical (e.g. resveratrol — potent antioxidant found in grapes). Here, we hypothesize that the potential therapeutic effect of food items depends on the diversity and availability of disease-beating bioactive molecules.
Hence, using publicly available databases of foods and bioactive molecules within foods, we built a food map with the theoretical anti-COVID-19 capacity of each ingredient ranked according to an enrichment score reflecting bioavailability and diversity of the predicted bioactive molecules with antiviral properties. We refer to these foods as antiviral hyperfoods.
Image source: Laponogov et al. — link
The top-ranked antiviral hyperfoods include different berries (blackcurrant, cranberry and blueberry), cruciferous vegetables (cabbage, broccoli), apples, citrus fruits (sweet orange and lemon), onions, garlic and beans.
Interestingly, blackcurrant has been shown to have not only potential antiviral effect, but also potential disinfectant and antiseptic effects against influenza  — although its ability to protect specifically against COVID-19 is yet to be evaluated. Similarly, a recent study highlighted the potential of cabbage and fermented vegetable consumption in minimizing adverse outcomes in COVID-19 .
Using machine-learning-based techniques, we identified bioactive molecules within foods with anti-COVID-19 properties based on their commonality at the genome level with experimental anti-COVID-19 drugs. We then built a food mapranking foods based on the bioavailability and diversity of the predicted antiviral molecules.
Even though we’re still some steps away from designing recipes that would be both tasty and healthy, our results show promise for the development of a clearer path for the management of non-severe cases of COVID-19, in which antiviral hyperfoods would be ‘prescribed’ to help patients recover faster and better.
. . .
I am grateful to Ivan Laponogov for proof-reading this post. Results presented here are part of a theoretical work, in process of clinical validation, and do not intend to substitute certified medical advice. This work is part of the project Hyperfoods, based at Imperial College, led by Kirill Veselkov in collaboration with the Vodafone Foundation and Kitchen Theory.
 Laponogov I*, Gonzalez G*, Shepherd M. et al. (2021). Network Machine Learning Maps ’Hyperfoods’ to Fight COVID-19. Human Genomics 15(1): 1.* equal contribution.
 ES Knock, LK Whittles, JA Lees et al. The 2020 SARS-CoV-2 epidemic in England: key epidemiological drivers and impact of interventions. Imperial College London (22–12–2020), doi: https://doi.org/10.25561/85146
 Ikuta K, Mizuta K, Suzutani T. Anti-influenza virus activity of two extracts of the blackcurrant (Ribes nigrum L.) from New Zealand and Poland. Fukushima J Med Sci. 2013;59(1):35–8. doi: 10.5387/fms.59.35.
 Bousquet J, et al. Cabbage and fermented vegetables: from death rate heterogeneity in countries to candidates for mitigation strategies of severe COVID-19. Allergy. 2020;00:1–16. https://doi.org/10.1111/all.14549.