Can Invalid Bioactives Undermine Natural Product-Based Drug Discovery?

Abstract

High-throughput biology has contributed a wealth of data on chemicals, including natural products (NPs). Recently, attention was drawn to certain, predominantly synthetic, compounds that are responsible for disproportionate percentages of hits but are false actives. Spurious bioassay interference led to their designation as pan-assay interference compounds (PAINS). NPs lack comparable scrutiny, which this study aims to rectify. Systematic mining of 80+ years of the phytochemistry and biology literature, using the NAPRALERT database, revealed that only 39 compounds represent the NPs most reported by occurrence, activity, and distinct activity. Over 50% are not explained by phenomena known for synthetic libraries, and all had manifold ascribed bioactivities, designating them as invalid metabolic panaceas (IMPs). Cumulative distributions of ∼200,000 NPs uncovered that NP research follows power-law characteristics typical for behavioral phenomena. Projection into occurrence-bioactivity-effort space produces the hyperbolic black hole of NPs, where IMPs populate the high-effort base.

Figure 1

Relationship among the source, the bioassay, and the interpretation of data from the NP discovery pathway is complex by nature.

Figure 2

Venn diagram of the three considered sets of the top 39 metabolites: most occurring (O), most reported as tested for activity (A), and most distinct (D) activities. The activities/distinct activities set are highly overlapping, whereas the occurrence set tends to be isolated from these.

Figure 3

Scatter plot for the merged sets of the 39 metabolites that are highly occurring (O) and have the most reported activities (A) of all nearly 200,000 NPs included in NAPRALERT with annotations matching the metabolite numbers in Table 2. While not showing a clear correlation between these two sets (A vs O), some metabolites are clearly outliers (1, 2, 3), and two major groups emerge for metabolites highly occurring and metabolites with a high number of activities reported.

Figure 4

Cumulative sums of the distributions of the three top-20 sets of NPs: occurrences (O) in blue, activities (A) in red, and distinct activities (D) in brown. The arrows show the percentage of citations at each given point, as well as the number of citations (in parentheses) that represent the top-20 NPs (bottom left), as well as the top 10% (lower left), top 50% (middle), and top 90% (upper right) of all NPs. The stars indicated the beginning of the single citation per compound zone, which continues on the right, ending at 100%.

Figure 5

Truncated power-law fitting of the distributions (blue) and cumulative sums (red) of the three sets: occurrences (O), activities (A), and distinct activities (D). These graphics represent the cumulative complementary density functions, representing the probabilities (y-axis) of obtaining a given value (x-axis). They clearly show that low-citation compounds (left) are more likely to happen than high-citations ones (right). Dotted green lines are the truncated power-law fitting according to eq 1.

Figure 6

Cumulative abundances of the reporting of the occurrences, activities, and distinct activities all follow the same principal power-law distribution. A typical curve is shown in A, indicating the two major regions of overattention to a few and lack of effort on many NPs. Distributing this NP–abundance–bioactivity space, which was built on the base of NAPRALERT’s nearly 200,000 compounds, into the third dimension (B) generates a hyperbolic structure that resembles a well-known corpus in astrophysics and is, therefore, termed the black hole of NPs. Panel C shows its various zones that categorize all NPs by their attached biological knowledge and abundance of the test parameter (O/A/D; see main text). Similar to a stellar black hole, density (representing research effort) increases dramatically toward the bottom (with infinite effort not being a scientific option). In distinction to its true counterpart, the black hole of NPs has a (virtual) outlet toward the bottom (C), which release either precious hits or IMPs.