Nat-UV DB: A Natural Products Database Underlying of Veracruz-Mexico

Database construction and curation

The database of natural products from the state of Veracruz was assembled from a literature search. For the construction of the first version of NAT-UV DB, PubMed, Google Scholar, Sci-Finder, Redalyc, and the institutional repository of the Universidad Veracruzana (Mexico) databases were searched using the keywords “natural product”, “NMR”, and “Veracruz.” We collected information from research articles, and bachelor, master, and doctorate theses from universities and research centers. To complement the data mining, two additional criteria were used for the final selection of the literature used to construct the database. The first filter was that the elucidation of the reported chemical structures has been supported by nuclear magnetic resonance (NMR). The second one was that the compounds identified were obtained from a natural source from any region in the state of Veracruz (Mexico). The search was generated by publication year from 1970 to June of 2024. We want to emphasize that this is the first version of Nat-UV DB; future versions will have natural products from more years, and more research repositories, to assemble a database representative of the entire biodiversity of the state of Veracruz. For each collected molecule, their isomeric SMILES strings²⁹ were generated with ChemBioDraw Ultra V.13, maintaining the stereochemistry reported in the primary literature.³⁰ With the module’Wash’, from the molecular operating environment (MOE) program, version 2024,³¹ the database was curated, maintaining without changes the stereochemistry reported of each molecule. This was done to normalize and collect the most relevant information from the molecules. The data curation involved the elimination of salts, the adjustment of the protonation states, and the elimination of the duplicated molecules. The default settings of the ‘Wash’ module were used. The information collected for each identified compound is organized according to the natural origin of its place of collection, like kingdom, genera, species, and geographical collection. Finally, the list of curated compounds was manually cross-referenced to PubChem³² and ChEMBL v.34³³ databases, which enabled the annotation of databases with the bioactivities that have been associated with each chemical structure ( Figure 2). For those compounds reported in theses, and which were evaluated in a biological test, the biological activity was also included in the database.

Figure 2. Workflow used to construct the Nat-UV database.

Reference data sets

In order to characterize the chemical diversity of Nat-UV DB and to explore its chemical space coverage, approved drugs³⁴ and the Latin American natural products compound database (LaNAPDB)³⁵ were used to compare their chemical structures and properties. The structure files used in this work were taken from open repositories of previously published analyses of natural products databases.³⁶ The structures of the reference compounds were curated using the same procedure described to prepare Nat-UV DB. Table 1 summarizes Nat-UV DB and the reference databases and the number of compounds. Of note, the reference collections included data sets of natural products, including two from Mexico.

Druglikness profiling

The curated Nat-UV DB database was characterized by calculating six physicochemical properties of pharmaceutical interest, namely: molecular weight (MW), octanol/water partition coefficient (ClogP), topological surface area (TPSA), number of rotatable bonds (RB), number of H-bond donor atoms (HBD), and number of H-bond acceptor atoms (HBA). The statistical analysis was done with the program DataWarrior v.06,³⁸ by calculating the mean, median, and standard deviation of the calculated properties. Based on these statistics Nat-UV DB was further compared with other databases (LANAPDB, BIOFACQUIM, UNIIQUIM, and approved drugs from DrugBank) ( Table 1). The systematic comparison was generated using the Python programming language. The code is freely available at https://github.com/EdgL2/Nat-UV-DB.

Scaffold content analysis of Nat-UV DB

The most frequent and unique molecular scaffolds of Nat-UV DB and reference databases ( Table 1) were computed using the scaffold definition of Bemis and Murcko.³⁹ This analysis was done using Python, the code of which is freely available at https://github.com/EdgL2/Nat-UV-DB.

Visualization of the chemical space

In order to generate a visual representation of the chemical space of Nat-UV DB, the fingerprint ECFP4 (1024 bits) was calculated for each compound⁴⁰ and the visualization was done using t-distributed stochastic neighbor embedding (t-SNE).⁴¹ The selection of this visualization method was based on recent studies that support its utility for the systematic study of small and large datasets in terms of neighborhood preservation and visualization capabilities.⁴² The ECFP4 fingerprint and the t-SNE coordinates were calculated in KNIME software. The optimization parameters we used in t-SNE were dimensions (3), iterations (10,000), theta (0.3), perplexity (30.0), and number of threats (8), using 28 as the seed number. The interactive visualization was implemented using DataWarrior software, version 06.⁴³ The KNIME workflow and data generated are freely available in the Software availability section.

Chemical diversity analysis

To compare the chemical diversity of Nat-UV DB with the reference data sets, we employed the consensus diversity (CD) plot which is a simple two-dimensional graph that helps to visualize and compare the diversity of several compound data sets considering multiple representations such as chemical scaffolds, and fingerprint-based diversity.⁴⁴ In this study, the CD plot was generated using the median paired similarity (ECFP4-1024 bits)/Tanimoto; x-axis) and the median paired scaffold similarity (Bemis-Murck representations using ECFP4-1014 bits/Tanimoto; y-axis).⁴⁴ Both are established and are representative metrics of the scaffold and fingerprint-based diversity.⁴⁵ Subsets of the compounds were retrieved from control data sets ( Table 1). The workflow implemented in KNIME software is available in the Software availability section.

Nat-UV database

As described in the Methods section, the scientific papers and thesis that complied with the inclusion criteria were selected. Each of the 45 scientific documents selected (1 Doctorate thesis degree; 8 Master thesis degrees; 36 research articles) was analyzed individually to extract manually the chemical structures of each identified natural product. The Nat-UV DB contains information that allows identifying the bibliography precedence of the data. For example: compound name, reference, digital object identifier (DOI), and publication year. Also, it contains data related to the natural source precedence of the data. For example: kingdom, genus, species, and geographical location of the collection of the natural source. Additionally, we added cross-referenced IDs with other databases (e.g. PubChem and ChEMBL). Finally, we manually cross-referenced each compound with their reported bioactivity contained in ChEMBL v. 34.

The current version of Nat-UV DB has 227 compounds collected from different geographical zones of Veracruz ( Figure 3A), mainly isolated by different kinds of gender plants ( Figure 3B). For example, the gender Hyptis, Capsicum, Nidema, Dryopteris, Ipomoea, Azadirachta, Hamelia, Croton, and Guarea are examples of the most frequently studied. Other species from other kingdoms also stand out as Aspergillus, Ganoderma, Colletotrichum, and Aegiale ( Figure 3C). Figure 1D illustrates the distribution of compounds per year reported since 1970 to date. Finally, 79% of the compounds contained in this database have been associated with almost one bioactivity report ( Figure 2E, D) which highlights compounds with anticancer and antimicrobial (antibacterial or antifungal) activities.

Figure 3. Descriptive analysis of the Nat-UV DB.

(A) Geographical collection of natural resources studied in this work, and the number of compounds obtained by each region; (B) Quantification of compounds contained in this database by genus; (C) Quantification of compounds contained in this database by specie precedence; (D) Number of isolated compounds by decades; (E) Associated bioactivity for the compounds contained in this database; and (F) Multi-activity landscape of compounds contained in this database.

Molecular scaffolds

From the total number of compounds contained in Nat-UV DB (227 compounds), 112 scaffolds were identified, of which 52 (52/112; 46%) are unique. Namely, Nat-UV DB contains scaffolds (52) that have not been reported previously in other Latin American datasets, and that are not present in the scaffolds collection of approved drugs ( Figure 4A). The most representative unique scaffolds of Nat-UV DB are shown in Figure 4B, highlighting the presence of derivatives of limonoids, butyrolactones, flavones, pentacyclic triterpenes, etc. The full list of unique scaffolds is available in the Data availability section.

Figure 4. Unique scaffold content in Nat-UV DB.

(A) Shared scaffolds of Nat-UV DB and reference datasets for natural products (LANaPDB, BIOFACQUIM, and UNIIQUIM) and approved drugs (Drugbank); (B) Representative unique scaffolds contained in Nat-UV DB.

There are previous reports of two Mexican natural products databases ( Table 1), but the BIOFACQUIM database is the unique one associated with collected geographical data. Interestingly, 74 compounds contained in this dataset were collected in the state of Veracruz (Mexico). This explains that 32 (32/112; 28%) scaffolds are shared in both databases ( Figure 4A). Also, 17 (17/112; 15%) scaffolds are shared between Nat-UV DB and UNIIQUIM, while 53 (53/112; 47%) scaffolds are shared between Nat-UV DB and the LaNaPDB. Finally, 24 (24/112; 21%) scaffolds were shared between Nat-UV DB and the approved drugs collection. In other words, Nat-UV DB contains some natural product scaffolds (60/112; 54%) that have been identified previously in Mexico and other Latin American countries or have been used as a drug.

Molecular properties

Figure 5 shows a violin plot of the distribution of the six drug-likeness properties calculated for Nat-UV DB. The distribution of the same properties for the two references used in this work was included in comparing the violin plots. ( Table 1). Intrinsic molecular properties like size, flexibility, and polarity are described by explicit molecular properties like weight (MW), coefficient of octanol/water partition (CLogP), number of H-acceptor and H-donors bonds, polar surface area (PSA), and number of rotatable bonds (RB) ( Figure 5A). Summary statistics are presented at the bottom of the violin plots ( Figure 5B).

Figure 5. Violin plots for the drug-likeness physicochemical properties of Nat-UV DB and reference data sets.

(A) The boxes inside of violins enclose data with values within the first and third quartile; (B) Summary statistics are included below each l plot. MW: molecular weight; ClogP: octanol/water partition coefficient; H-bond acceptors: number of H-bond acceptor atoms; H-Donors: number of H-bond donor atoms; PSA: polar surface area; RB: number of rotatable bonds.

According to Figure 5 the size (MW, HA, and HB), flexibility (RB), and permeability (PSA) profiling of Nat-UV are comparable with the control datasets. However, the polarity (CLogP) of the compounds contained in Nat-UV DB, LANaPDB, BIOFACQUIM, and UNIIQUIM is higher than the approved drugs, however, the Nat-UV DB exhibited a shorter distribution than each natural products databases. This finding agrees with previous reports indicating that natural products are slightly more hydrophobic than drugs approved for clinical use.³⁶

Chemical space and diversity analysis

Figure 6 shows a visual representation of the chemical space of Nat-UV DB based on ECFP4 fingerprint using t-SNE. Figure 6(B-E) compares Nat-UV DB with other natural products databases (i.e. LANaPDB, BIOFACQUIM, and UNIIQUIM) and approved drugs. Interestingly, Nat-UV DB shares part of its chemical space with the approved drugs dataset, but Nat-UV DB compounds are more distributed in the three dimensions of the plot, which suggests that have a higher structural diversity than the approved drug dataset. However, LANaPDB (the largest dataset analyzed in this study) has an apparently higher structural diversity than the other studied datasets. To quantify the diversity of each dataset, the calculation of structural diversity and scaffold diversity were done ( Figure 6F). To quantify the diversity of each dataset, we calculated the mean of the paired similarity of the structures (x-axis) and scaffolds (y-axis) based on the similarity of each pair of compounds in the dataset using ECFP4 fingerprint and Tanimoto coefficient, where the higher values confirm a higher structural or scaffold diversity of the dataset. These results showed that Nat-UV DB has a higher structural and scaffold diversity than the approved drugs. However, it has low structural and scaffold diversity in contrast with UNIIQUIM and LANaPDB. Finally, Nat-UV DB shows a higher structural diversity than BIOFACQUIM, but a lower scaffold diversity than this one.

Figure 6. Visual representation of the chemical space coverage of Nat-UV DB and reference datasets based on ECFP4 and t-SNE as a visualization method.

(A) Nat-UV DB; (B) Nat-UV DB and Drugbank (approved drugs); (C) Nat-UV DB and LANaPDB; (D) Nat-UV DB and BIOFACQUIM; (E) Nat-UV DB and UNIIQUIM; and (F) Consensus diversity plot of Nat-UV DB and the four reference datasets.