Case Studies


Drug Space

This ‘drug space’ was generated using StarDrop’s chemical space visualisation using a set of 1395 marketed small molecule drugs. In this space, the proximity of two points represents the structural similarity between the corresponding compounds. This provides a convenient way to map the distribution of compounds or their properties across the chemical diversity of drugs. The figure below shows some illustrative structures for different regions of the space.

Technical Details

In generating this space, the similarity between two compounds is defined using a Tanimoto index based on a 2D path-based fingerprint. The distribution of points is generated using the t-distributed stochastic neighbour embedding algorithm [1].

Data Set

The compound structures used to create the drug space were downloaded from the ChEMBL approved drug list [2]. Only compounds assigned a ‘Development Phase’ of 4 were retained. Duplicate structures and compounds with molecular weight less than 100 Da or greater than 1000 Da were removed.

StarDrop users can download a Drug Space StarDrop Project containing this data set and plot their compounds into the same drug space.

Pharmacokinetics Scoring Profile

Each of the top-10 cardiovascular (CV) compounds were scored using StarDrop’s Probabilistic Scoring approach [3] for multi-parameter optimisation. This assesses the overall balance of properties against a profile of criteria representing the desired properties of a high quality compound for the therapeutic objectives of a project. The overall score, between 0 and 1, represents the likelihood of achieving the ideal outcome for all properties. The top-10 cardiovascular drugs were scored based on their clinically observed pharmacokinetic (PK) properties:

  • Oral bioavailability
  • Half-life (h)
  • Plasma-protein binding (%)
  • Volume of distribution (L)

For each compound, a histogram shows the performance of each of these properties against the desired property values, as illustrated right.

A high bar for a property indicates a good value of the property while a low bar corresponds to a poor value against an important property criterion. The data for each of the compounds, along with references to the sources, are provided in the table in the Data section below.

The scoring profile used to compare the top-ten CV compounds was as follows:

The range of desired values for each of the properties is shown above, along with the relative importance of each property criterion. However, the criteria are not defined as hard cut-offs, which could draw artificially harsh distinctions between compounds with similar property values near to the boundaries of the desired ranges. Instead, a ‘desirability function’ has been defined for each criterion, relating the value of each property to its desirability. The desirability functions used in this profile are shown below (in each case the desirability function is shown in blue and the histogram shows the distribution of the corresponding property values for the top-10 CV compounds):

This scoring profile is included in a CV top-10 StarDrop project which StarDrop users can download to score their own compounds, or modify as required.

The choice of a property profile is subjective to some degree, but the rationales for the choice of these desirability functions are as follows:

    • For an orally dosed compound, high oral bioavailability is desirable to achieve good systemic exposure. Typically, a value above 50% would be ideal, but below this, the higher the oral bioavailability the better. A value of zero is clearly unacceptable.
    • A half-life between 8 and 16 hours would be suitable for twice-daily dosing. A half-life below this would increase the likelihood of requiring a larger number of doses or a larger dose to maintain a therapeutic concentration. Requiring a higher number of doses increases the risk of poor patient compliance, while administration of large doses would require a high therapeutic index. A half-life above 16 hours increases the risk of accumulation, potentially leading to toxicity. This was considered a lower risk, so a high half-life is given a lower penalty than a low half-life in the desirability function for this property.
    • High plasma-protein binding reduces the free concentration of a compound; therefore, plasma-protein binding values close to 100% are undesirable. However, very low plasma-protein values are difficult to achieve; therefore, a desired range of less than 90% is defined.
    • A volume of distribution above 70L indicates distribution within total body water, hence a value greater than this would be desirable, with an ideal range between 210L and 350L, indicating good tissue exposure. A much higher volume of distribution indicates a greater degree of non-specific binding, increasing the risk of toxicity or undesirable pharmacology; therefore, values above 350L are assigned decreasing desirability, with values above 490 L given the lowest desirability.

Of course, there is no single ‘ideal’ pharmacokinetic profile for a CV drug and there are certainly exceptions to these ‘rules’. For example, Simvastatin is a component of Vytorin and has very low oral bioavailability. However, Simvastatin is a pro-drug and is converted almost entirely to the active beta-hydroxy form by first-pass metabolism. Therefore, poor oral bioavailability is acceptable for the parent because exposure to the active metabolite is good.

Drug Data

Sales Rank

Product Name

Company Name

Active Ingredient


Oral bioavailability (%)

Volume of distribution (L)

Plasma-protein binding (%)

Half-life (h)

Sales ($M)[1]



AstraZeneca Rosuvastatin Calcium HMG-CoA reductase inh.

20 [4]

134 [4]

88 [4]

19 [4]




Merck & Co Ezetimibe Intestinal abs. inh.

93 [5]

105 [6]

90 [5]

22 [7]




Sanofi Clopidogrel Bisulfate [2] Irreversible P2Y12 inh.

>50 [8]

33 [9]

98 [8]

8 [8]




Pfizer Inc. Atorvastatin Calcium HMG-CoA reductase inh.

14 [10]

381 [10]

98 [10]

14 [10]




Merck & Co Ezetimibe and Simvastatin [3] HMG-CoA reductase inh.

0 [22]

116 [11]

95 [12]

3.5 [11]




AstraZeneca Ticagrelor


Reversible P2Y12 inh.

36 [13]

116 [13]

99.7 [13]

7 [13 ]




Gilead Sciences Ranolazine Unknown

76 [14]

132.5 [15]

62 [14]

7 [14]




Bayer Acetylsalicylic Acid COX-1 & COX-2 irrev. Inh.

68 [16]

10.5 [17]

99.5 [18]

0.25 [18]




Eli Lilly Prasugrel [4] Irreversible P2Y12 inh.

79 [19]

56 [20]

98 [20]

7.4 [20]




Sanofi Dronedarone Na/K/Ca channel inh.

4 [21]

1400 [5] [21]

98 [21]

16 [21]


These data can be downloaded in a CV top-10 StarDrop project along with the compound structures and scores.

1 Sales data obtained from PHARMACOMPASS,
2 Clopidogrel is a prodrug. Therefore, the half-life and volume of distribution are presented for the active metabolite.
3 Pharmacokinetic and target data shown for Simvastatin. Data for Ezetimibe is shown above.
4 Pasugrel is a prodrug and is rapidly and extensively converted into its active metabolite. Therefore, PK data is presented for the active metabolite.
5 The intravenous volume of distribution is shown.


    • Zetia (Ezetimibe) is absorbed and conjugated to a phenolic glucuronide which is the subject of enterohepatic recirculation. As an intestinal absorption inhibitor, this increases the exposure at the site of action and leads to a prolonged half-life.
    • Plavix (Clopidogrel) is a prodrug and is converted into its active carboxylic acid metabolite by cytochrome P450 metabolism. For this reason, the half-life and volume of distribution of the active metabolite are shown because they are more relevant to its efficacy.
    • Vytorin is a combination therapy of Ezetimibe and Simvastatin. Ezetimibe is represented individually in the top-10; therefore, the PK data for Vytorin represents the properties of Simvastatin. Simvastatin is a pro-drug and is extensively metabolized into the active beta-hydroxy metabolite on first pass; therefore, the low oral bioavailability of the parent compound is, in practice, not a liability for this drug.
    • Effient (Prasugrel) is a prodrug and is extensively metabolised on first-pass to form its active metabolite; therefore, the PK data for the active metabolite are shown because they are more relevant to its efficacy.


1 van der Maaten LPJ, Hinton GE. Visualizing High-Dimensional Data Using t-SNE. Journal of Machine Learning Research. 2008;9:2579-2605.

2 European Bioinformatics Institute. ChEMBL Approved Drugs. [Internet]. [cited 2016 August 28]. Available from:

3 Segall MD. Multi-Parameter Optimization: Identifying high quality compounds with a balance of properties. Curr. Pharm. Des. 2012;18(9):1292-1310.

4 DrugBank. Rosuvastatin. [Internet]. [cited 2016 August 28]. Available from:

5 Drugs at FDA. Zetia NDA. [Internet]. [cited 2016 August 28]. Available from:

6 Drug Information System. Ezetimibe. [Internet]. [cited 2016 August 28]. Available from:

7 DrugBank. Ezetimibe. [Internet]. [cited 2016 August 28]. Available from:

8 DrugBank. Clopidogrel. [Internet]. [cited 2016 August 28]. Available from:

9 Yousef AM, Diab O, Arafat T, Khribash A, El-Saleh AAHH. Pharmacokinetics of 600 mg loading dose of clopidogrel in patients undergoing percutaneous corony intervention. African J. Pharm. and Pharmacol. 2013;7(11):574-584.

10 DrugBank. Atorvastatin. [Internet]. [cited 2016 August 28]. Available from:

11 Kosoglou Tea. Pharmacodynamic interaction between the new selective cholesterol absorption inhibitor ezetimibe and simvastatin. Br. J. Clin. Pharamacol. 2002;54(3):309-319.

12 DrugBank. Simvastatin. [Internet]. [cited 2016 August 28]. Available from:

13 DrugBank. Ticagrelor. [Internet]. [cited 2016 August 28]. Available from:

14 DrugBank. Ranolazine. [Internet]. [cited 2016 August 28]. Available from:

15 Jerling M. Clinical Pharmacokinetics of Ranolazine. Clin. Pharmacokinet. 2006;45(5):469-491.

16 Needs CJ, Brooks PM. Clinical pharmacokinetics of aspirin. Clin. Pharamacokinet. 1985;10(2):164-77.

17 University de Lausanne. Aspirin. [Internet]. [cited 2016 August 28]. Available from:

18 DrugBank. Acetylsalicylic acid. [Internet]. [cited 2016 August 28]. Available from:

19 Drugs at FDA. Prasugrel prescribing information. [Internet]. [cited 2016 August 28]. Available from:

20 DrugBank. Prasugrel. [Internet]. [cited 2016 August 28]. Available from:

21 DrugBank. Dronedarone. [Internet]. [cited 2016 August 28]. Available from:

22 Drugs at FDA. Vytorin prescribing information. [Internet]. [cited 2016 August 28]. Available from: