PhyteByte: identification of foods containing compounds with specific pharmacological properties

Abstract

Background: Phytochemicals and other molecules in foods elicit positive health benefits, often by poorly established or unknown mechanisms. While there is a wealth of data on the biological and biophysical properties of drugs and therapeutic compounds, there is a notable lack of similar data for compounds commonly present in food. Computational methods for high-throughput identification of food compounds with specific biological effects, especially when accompanied by relevant food composition data, could enable more effective and more personalized dietary planning. We have created a machine learning-based tool (PhyteByte) to leverage existing pharmacological data to predict bioactivity across a comprehensive molecular database of foods and food compounds. Results: PhyteByte uses a cheminformatic approach to structure-based activity prediction and applies it to uncover the putative bioactivity of food compounds. The tool takes an input protein target and develops a random forest classifier to predict the effect of an input molecule based on its molecular fingerprint, using structure and activity data available from the ChEMBL database. It then predicts the relevant bioactivity of a library of food compounds with known molecular structures from the FooDB database. The output is a list of food compounds with high confidence of eliciting relevant biological effects, along with their source foods and associated quantities in those foods, where available. Applying PhyteByte to the human PPARG gene, we identified irigenin, sesamin, fargesin, and delta-sanshool as putative agonists of PPARG, along with previously identified agonists of this important metabolic regulator. Conclusions: PhyteByte identifies food-based compounds that are predicted to interact with specific protein targets. The identified relationships can be used to prioritize food compounds for experimental or epidemiological follow-up and can contribute to the rapid development of precision approaches to new nutraceuticals as well as personalized dietary planning.

Predicting health effects of food compounds via ensemble machine learning

“…Machine learning techniques have demonstrated advantages in unravelling the quantitative structureactivity relationships (QSAR) between food chemicals and health effects (Frenzel et al, 2017;Westerman et al, 2020;Gonzalez et al, 2021). Gonzalez et al (2021) adopt graph neural networks to learn representations of chemical molecules, based on which to train a multi-layer perceptron (MLP) on drugs to predict anticancer bioactivities of food compounds.…”

“…The results show that the knowledge learnt from drugs is transferable to food molecules. Westerman et al (2020) use EP2 fingerprints of drug molecules to train a random forest model for predicting the pharmacological effects of food compounds. Besides health benefits, the knowledge about drug side effects has also been exploited to predict health risks of food chemical contaminants such as mutagenicity and carcinogenicity (Frenzel et al, 2017).…”

Predicting health effects of food compounds via ensemble machine learning

“…Arti cial intelligence especially machine learning techniques are supposed to demonstrate full advantages in unravelling quantitative structure-activity relationship (QSAR) of food chemical compounds and gaining molecule-level insights into their health and nutritional effects. At present, the studies of foodomics data mining focus on avor compounds [29,30], bene cial or adverse health effects of food ingredients [31][32][33][34] and bioactive peptides [35,36], food-drug interactions [37,38], food compounds identi cation [39] and metabolic pathways of food compounds [40]. Bi et al [30] rst convert GC-MS spectrum data into ngerprint images and then learn the ngerprint representations of compounds via convolutional neural networks (CNN) for food avor evaluation.…”

“…For instance, Gonzalez et al [32] use graph neural networks to learn the vector representations of drugs and then train a multi-layer perceptron (MLP) to predict anticancer bioactivity of food molecules. The work [33] use drugs' EP2 ngerprints to train a random forest model for predicting pharmacological effects of food compounds. Beside health bene ts, the knowledge about drug side-effects is also exploited to predict health risks such as mutagenicity and carcinogenicity of food chemical contaminants [34].…”

A Catalog of Natural Products Occurring in Watermelon—Citrullus lanatus

“…Regarding human nutrition, knowledge of what is in a food is the basis by which to characterize the health benefits of that food. Those efforts support knowing what to eat to remain healthy ( 12 ) and assist in defining the “dark matter” or chemical complexity of nutrition ( 13 , 14 ). In addition, comprehensive catalogs of the biochemicals present in a crop can stimulate projects in plant breeding and crop improvement, especially when coupled with genome sequencing and other such data streams ( 15 , 16 ).…”