Probing the links between in vitro potency, ADMET and physicochemical parameters

Abstract

A common underlying assumption in current drug discovery strategies is that compounds with higher in vitro potency at their target(s) have greater potential to translate into successful, low-dose therapeutics. This has led to the development of screening cascades with in vitro potency embedded as an early filter. However, this approach is beginning to be questioned, given the bias in physicochemical properties that it can introduce early in lead generation and optimization, which is due to the often diametrically opposed relationship between physicochemical parameters associated with high in vitro potency and those associated with desirable absorption, distribution, metabolism, excretion and toxicity (ADMET) characteristics. Here, we describe analyses that probe these issues further using the ChEMBL database, which includes more than 500,000 drug discovery and marketed oral drug compounds. Key findings include: first, that oral drugs seldom possess nanomolar potency (50 nM on average); second, that many oral drugs have considerable off-target activity; and third, that in vitro potency does not correlate strongly with the therapeutic dose. These findings suggest that the perceived benefit of high in vitro potency may be negated by poorer ADMET properties.

Figure 1

Relationship between ADMET parameters and physicochemical properties. (a) Plot of molecular mass versus AlogP for 1,791 oral drugs. The data are coloured according to the absorption, distribution, metabolism, excretion and toxicity (ADMET) score, which is a measure of the deviation from oral drug space as given by molecular mass and AlogP. (b) A comparison of the ADMET score distribution of oral drugs (coloured according to the score) and 201,355 unique compounds with measured target potency data from the ChEMBL database (coloured in black). Approximately 14% of oral drugs have a score >2 compared to ~39% for drug discovery compounds reported in ChEMBL.

Figure 2

Relationship between in vitro potency and physicochemical properties. The graphs illustrate the relationship between the mean reported in vitro potency (shown here as the negative logarithm of the XC₅₀ value; pXC₅₀) and molecular mass (a) or AlogP (b) for 201,355 compounds with reported activity measurements in the ChEMBL database. c | The effect of both parameters shows some independence as, for a given AlogP value, a concomitant increase in molecular mass increases the mean pXC₅₀ value and vice versa. Only 9 observations are present in the AlogP 4.0–5.0, molecular mass <200 category, hence the large error bars. Plotted values are offset within x-axis bins to aid visualization. Error bars denote the 95% confidence interval in the means

Figure 3

Relationship between promiscuity and physicochemical properties. The graphs illustrate the relationship between the mean promiscuity and molecular mass (a), AlogP (b) or ionization state (c) for a diverse set of 40,408 molecules in the ChEMBL database. Each compound has ≥3 activity measurements reported in ChEMBL. The total number of observations in each molecular mass bin, in order of increasing mass are: 924, 5,121, 12,287, 12,514 and 9,562. The corresponding values for AlogP are: 8,741, 6,839, 9,033, 7,632 and 8,163, respectively. The corresponding values for ionization state are: 22,060, 10,893, 5,862 and 1,593, respectively.

Figure 3

Relationship between promiscuity and physicochemical properties. The graphs illustrate the relationship between the mean promiscuity and molecular mass (a), AlogP (b) or ionization state (c) for a diverse set of 40,408 molecules in the ChEMBL database. Each compound has ≥3 activity measurements reported in ChEMBL. The total number of observations in each molecular mass bin, in order of increasing mass are: 924, 5,121, 12,287, 12,514 and 9,562. The corresponding values for AlogP are: 8,741, 6,839, 9,033, 7,632 and 8,163, respectively. The corresponding values for ionization state are: 22,060, 10,893, 5,862 and 1,593, respectively.

Figure 3

Relationship between promiscuity and physicochemical properties. The graphs illustrate the relationship between the mean promiscuity and molecular mass (a), AlogP (b) or ionization state (c) for a diverse set of 40,408 molecules in the ChEMBL database. Each compound has ≥3 activity measurements reported in ChEMBL. The total number of observations in each molecular mass bin, in order of increasing mass are: 924, 5,121, 12,287, 12,514 and 9,562. The corresponding values for AlogP are: 8,741, 6,839, 9,033, 7,632 and 8,163, respectively. The corresponding values for ionization state are: 22,060, 10,893, 5,862 and 1,593, respectively.

Figure 4

Relationship between promiscuity and ionization state. The data shown are for a diverse set of 40,408 molecules and are broken down by molecular mass to aid analysis. Error bars denote the 95% confidence interval in the means. Plotted values are offset within x-axis categories to aid visualization.

Figure 5

Relationship between in vitro potency and dose for oral drugs. Distribution of the lowest reported oral dose for 792 oral drugs (a) and the maximum reported in vitro pXC₅₀ for a subset of 261 drugs for which the target(s) are known and relevant potency data are available. The distribution of the minimum, mean and maximum doses per compound have median values corresponding to 63 μmol, 125 μmol and 223 μmol, or 24 mg, 47 mg and 83 mg, respectively. The corresponding median values for the minimum, mean and maximum therapeutically relevant pXC₅₀ values are 7.0, 7.3 and 7.7, respectively. c | Relationship between the reported mean therapeutic dose and the mean therapeutically relevant pXC₅₀ for the set of 261 oral drugs. The relationship between therapeutic dose and pXC₅₀ is displayed using a regression plot rather than analysis of variance (ANOVA), as it is a comparatively strong relationship. The correlation between therapeutic dose and molecular mass has a correlation coefficient of r² = 0.10, whereas the corresponding value between molecular mass and pXC₅₀ is r² = 0.16.

Figure 6

Promiscuity of oral drugs. The number of hits ≤1 μM reported for a subset of 392 oral drugs extracted from ChEMBL. The percentage of compounds with the number of reported hits indicated are shown next to each portion of the pie chart. This is likely to underestimate the promiscuity of oral drugs given the scarcity of biological data per compound in ChEMBL.