(C) PLOS One [1]. This unaltered content originally appeared in journals.plosone.org.
Licensed under Creative Commons Attribution (CC BY) license.
url:https://journals.plos.org/plosone/s/licenses-and-copyright

------------



Bioactivity assessment of natural compounds using machine learning models trained on target similarity between drugs

['Vinita Periwal', 'European Molecular Biology Laboratory', 'Heidelberg', 'Medical Research Council Toxicology Unit', 'University Of Cambridge', 'Cambridge', 'United Kingdom', 'Stefan Bassler', 'Faculty Of Biosciences', 'Heidelberg University']

Date: 2022-05

A catalogue of 11,788 natural compounds was obtained from FooDB ( www.foodb.ca ) ( S1C Table ). These correspond to 261 unique food sources and are categorized into 15 main food types such as vegetables, fruits, herbs and spices, and milk products ( S1B Fig ). For the simplicity in representation in S1B Fig , the frequency accounts for only one source per compound; however, a particular compound can be present in multiple food sources. The food compounds were structurally classified into 21 classes (see Material and methods ) ( S1B Fig ). Highly represented were lipids and lipid-like molecules (4803), phenylpropanoids and polyketides (2476), organoheterocyclics (1381) and organic oxygen compounds (1120). All these natural compounds were used to create an assessment library, where each chemical pair comprised of a drug and a natural compound ( Fig 1A ) (now referred to as drug-food pair). Pairwise similarities between each natural compound ( S1C Table ) and all the drugs were computed using the same set of predictors (i.e., using same predictors molecular fingerprints, MCS and molecular descriptors) as was for training dataset.

Similarity predictions by RF model.

The similarity of each of the 11k natural compounds paired to each of the 1410 drugs was evaluated using the trained RF models. Since all 5 split sets performed optimally, we chose to accommodate the drug-food similarity predictions from all 5 RF models. As many drugs have originated from natural compounds, drug-food pairs with a very high similarity fingerprint score (i.e., tanimoto score > 0.9) were removed (n = 1,850 pairs) considering them to be the same compound. Overall, the number of drug-food pairs compared on each RF model were 1,941,762. Pairs which passed the threshold of 0.5 probability are considered as a match. The number of hits with the 5 RF models are shown in Fig 2A. We picked drug-food pairs which were predicted as match by at-least 3 models (686 pairs) in further analysis. The full list of these 686 pairs is provided in S2B Table.

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 2. Drug-food compound similarity. (A) Number of hits retrieved from each split-sets model. (B) 200 drug-food pairs predicted as ‘match’ at the probability threshold of >0.5. The drugs are arranged according to their therapeutic class and food compounds according to their food source. The highlighted colored links represent the case examples in the five author defined groups (details in the text). (C) Group4-probable lead example taken up for experimental validation. The food compound 5-methoxysalicylic acid was a hit with the drug triflusal which has 4 known targets. We validated the inhibitory activity of triflusal and 5-methoxysalicylic acid against the target PTGS1 (also known as Cox-1). https://doi.org/10.1371/journal.pcbi.1010029.g002

These 686 pairs comprise of 329 unique food compounds and 289 unique drugs (full annotated list in S4B Table). Note that a drug can share similarity with more than one food compound and vice-versa. Also, a food compound can be present in multiple food sources, for exhaustive listing of known sources, we recommend querying the FooDB using respective compound Ids or names.

We performed manual curation of 30% of these 686 drug-food pairs (200 pairs) and categorized the food compounds into five custom defined groups based on the meta information available in the public domain (S4A Table and Fig 2B). Group 1, Analogs: food compounds which themselves represent a drug. Group 2, Endogenous: compounds reported as a metabolite in humans. Group 3, Experimental: food compounds currently under investigational or clinical trial as a therapeutic. Group 4, Probable lead: compounds with potentially novel bioactivity. Group 5, Others: compounds currently used in industrial application or used as additive or flavor enhancer. Each group is discussed below with case examples.

Group 1-analogs (56 pairs, 20 natural compounds)—food compounds in this group were found to be structural analogues to other known drugs i.e., the food compound paired with the drug is apparently another drug itself. This observation is consistent with the fact that the origin of many drugs is from natural sources. Yet, presence of some of these compounds in food is intriguing. A compound referred to as ‘Satiomem’ in FooDB (reported in barley and onions) resembles the drug Carbinoxamine, an antihistamine. It shared similarity with other antihistamines (such as Chlorprothixene, Doxepin, Antazoline and Chlorphenamine, Fig 2B, highlighted in blue) belonging to the nervous ATC category which are used as antipsychotics. These drugs share ‘Histamine H1 receptor’ as a target but there was no evidence found for Satiomem/Carbinoxamine having antipsychotic activity so it could potentially serve as a good candidate for further testing as an antipsychotic or resulting in drug interactions when used in combination with these drugs. This group also provides an opportunity to explore drug-repurposing.

Group 2-endogenous (49 pairs, 24 natural compounds)—compounds that are endogenous to human tissues but also reportedly present in various food sources. For example, ‘desoxycorticosterol’ a.k.a. 21-Hydroxyprogesterone was reported to be present in rice and is endogenously present in amniotic fluid and blood throughout human tissues. It’s predicted to be similar to other Hormonal and Genitourinary drugs (Fig 2B, highlighted in pink). ‘Estriol’, an estrogen produced by the human body, is reported to be present in pomegranate and beans.

Group 3-experimental (10 pairs, 5 natural compounds)—these food compounds are already under experimental investigation category (i.e., under approval to be used as drugs, reported in DrugBank accessed January 2018). ‘Higenamine’ is reported to be present in opium and coffee. This compound is in clinical trial (DrugBank id: DB12779) and has been patented for various therapeutic applications (Fig 2B, highlighted in purple). This group serve as a proof of principle that we could recall natural compounds with similar activity as currently used human-targeted drugs, which are being actively investigated pre-clinically.

Group 4-probable lead (76 pairs, 58 natural compounds)–to our knowledge, the compounds in this group have little or no hitherto reported evidence of their physiological or biological activity. The drug Papaverine is an alkaloid which is a vasodilator. ‘Annocherine B’ reportedly present in many fruits showed high similarity with Papaverine, however no evidence or reports of this compound about its action or use was found. ‘Bevantolol’ is a cardiovascular drug which shared similarity with two novel compounds ‘Codamine’ and ‘Laudanine’ (both reported in opium).

Group 5-others (9 pairs, 5 natural compounds)–food compounds found in this group are reported to be used as food additives such as flavor enhancers or have other industrial applications such as emulsifiers. ‘Neoisomenthol’ and ‘Isomenthol’ are used as a flavoring agent are similar with nervous category drugs ‘Codeine’, ‘Dezocine’, and ‘Tapentadol’. Thus, these five groups highlighted interesting similarity relationships existing between drug and food compounds and their wider therapeutic potential.

[END]

[1] Url: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1010029

(C) Plos One. "Accelerating the publication of peer-reviewed science."
Licensed under Creative Commons Attribution (CC BY 4.0)
URL: https://creativecommons.org/licenses/by/4.0/


via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.