Capturing nature's diversity

Abstract

Natural products are universally recognized to contribute valuable chemical diversity to the design of molecular screening libraries. The analysis undertaken in this work, provides a foundation for the generation of fragment screening libraries that capture the diverse range of molecular recognition building blocks embedded within natural products. Physicochemical properties were used to select fragment-sized natural products from a database of known natural products (Dictionary of Natural Products). PCA analysis was used to illustrate the positioning of the fragment subset within the property space of the non-fragment sized natural products in the dataset. Structural diversity was analysed by three distinct methods: atom function analysis, using pharmacophore fingerprints, atom type analysis, using radial fingerprints, and scaffold analysis. Small pharmacophore triplets, representing the range of chemical features present in natural products that are capable of engaging in molecular interactions with small, contiguous areas of protein binding surfaces, were analysed. We demonstrate that fragment-sized natural products capture more than half of the small pharmacophore triplet diversity observed in non fragment-sized natural product datasets. Atom type analysis using radial fingerprints was represented by a self-organizing map. We examined the structural diversity of non-flat fragment-sized natural product scaffolds, rich in sp3 configured centres. From these results we demonstrate that 2-ring fragment-sized natural products effectively balance the opposing characteristics of minimal complexity and broad structural diversity when compared to the larger, more complex fragment-like natural products. These naturally-derived fragments could be used as the starting point for the generation of a highly diverse library with the scope for further medicinal chemistry elaboration due to their minimal structural complexity. This study highlights the possibility to capture a high proportion of the individual molecular interaction motifs embedded within natural products using a fragment screening library spanning 422 structural clusters and comprised of approximately 2800 natural products.

Fig 1. PCA analysis of the DNP.

Comparison of fragment-sized natural products to all known natural products (DNP) in physicochemical property space as defined by principal component analysis of 11 physicochemical descriptors. The first three principal components account for 89.0% of the variance in the data (57.6%, 21.2%, and 10.2%, respectively). (a) PCA plot of 165281 clean natural products (145096 non-fragment-sized NPs (green) and 20185 fragment-sized NPs (blue). (b) PCA plot of 94929 natural products (74744 non-fragment sized natural products which satisfy Lipinski’s Ro5 (red) and 20185 fragment-sized NPs (blue)).

Fig 2. Pharmacophore analysis of natural products.

Based on the number of unique pharmacophore triplets (1–6 bonds), generated using eight features, we identified that fragment-sized natural products cover ~66% of the unique pharmacophore of the whole DNP.

Fig 3. Distribution of compounds within the SOMs trained using ECPF_4 fingerprints of 20185 fragment-sized natural products (Fig 3A and 3B) and 7365 non-flat fragment-sized natural products (Fig 3C–3H).

(a) 20185 fragment-sized natural products. (b) 7365 non-flat fragment sized natural products (Fsp ³* > 0.45). (c) 7365 non-flat fragment-sized natural products. (d) 1-ring molecules; 37% coverage of non-flat fragments. (e) 2-ring molecules; 68% coverage of non-flat fragments. (f) 3-ring molecules; 56% coverage of non-flat fragments. (g) 4-ring molecules; 21% coverage of non-flat fragments. (h) 5-ring molecules; 2% coverage of non-flat fragments. Each cell represents a cluster of fragments and the distance between cells (i.e. nearby cell are structurally related compounds) is indicated by the shading of the cell borders; darker borders indicate larger distance. Cells are coloured by population, with white for empty cells, and red for cell containing more than 5 compounds. The trained SOM is characterized by a toroidal architecture, which means that the top edge is connected to the lower edge and the left edge with the right edge.

Fig 4. Structural diversity analysis of 7365 non-flat fragment-sized natural products based on pharmacophore (orange) and radial (red) fingerprints.

Using SOMs (generated by training non-flat structures subset) we analyzed the distribution and therefore the coverage of the total diversity of non-flat fragments by the subsets clustered on the number of rings present in the structure.

Fig 5. Distribution of 2-ring fragments dataset within the non-flat fragment-sized natural products SOM.

For a few selected cells are, the most representative structures (green circle) is shown. In the black square box is reported one example of the molecular similarity within the same cell. The entire list of seed compounds can be found in S3 Table.