Drug Discovery & Pharmaceuticals Getting the Most Value from Your Compound Data Matthew Segall, Optibrium Ltd., 7221 Cambridge Research Park, Beach Drive, Cambridge, CB25 9TL, UK, Email: matt.segall@optibrium.com, Tel: 01223 815900, Fax: 01223 815907


In order to mitigate potential risks in drug discovery, data are routinely generated for large numbers of compounds across many properties at great effort and expense. However, much of these data do not have sufficient impact on compound design and selection decisions and too often they are consigned to a database without ever being considered. This unfortunate situation arises because decision-making based on complex, multi-parameter data is challenging. As a result there is a tendency to oversimplify and focus decisions around smaller numbers of better-known properties, discounting potentially valuable information.


This article explores ways in which greater value can be gained from the data generated in drug discovery laboratories. Intuitive multi-parameter optimisation approaches make it easy to include all relevant data in the decision-making process, guiding the selection of compounds with the best balance of properties for a successful drug, while ensuring that each property has an appropriate influence. The article will also illustrate how property data can be easily modelled, thereby capturing and visualising the relationships between chemical structures and their properties, to guide the optimisation of new compounds. If provided in a user- friendly, interactive way, that is accessible to all members of a project team, these approaches improve the efficiency and productivity of the drug discovery process.


Drug discovery is a risky business! The large majority of projects fail and, even for those that nominate a development candidate, only approximately one in twelve result in a drug that reaches the market [1]. This success rate has remained relatively unchanged over the last decade, but the cost per new chemical entity has increased dramatically from an estimated $800M in 2001 [2] to $1,800M in 2010 [1].


There are many causes of project failure, including changes in commercial priorities and lack of efficacy due to unvalidated targets, but a significant proportion are due to inappropriate physicochemical properties, pharmacokinetics (PK) or toxicity [1]. In an effort to address these causes of attrition, a wide range of properties are now routinely monitored in early drug discovery, using predictive models and experimental assays. For example, in the ‘90s, a high level of attrition in the clinic due to poor PK was noted [3], leading to the introduction of approaches for early measurement of absorption, distribution, metabolism and elimination (ADME) properties [4]. The result of this effort has been a reduction in the proportion of compounds failing in the clinic due to PK issues, from an estimated 39% in 1991 [3] to approximately 10% in 2001 [1]. However, the overall success rate in development remained almost constant over the same period [1] [5] because attrition due to toxicity issues has increased commensurately. In reaction to this, there is a current drive to measure more toxicity-related properties earlier in the drug discovery process.


The result of these trends is that many endpoints are now routinely measured for an increasing number of compounds. A successful drug must exhibit a delicate balance of these many properties and the quantity and complexity of these data make them difficult to use effectively.


Understandably, people find it difficult to make good decisions based on complex data, particularly when the data are uncertain and there is significant risk [6]. Therefore, there is a natural tendency to initially focus on a small number of parameters and consider other data later. However, this often means that much of the available data are not given sufficient consideration and their value is lost. Indeed, there are many anecdotes of data being consigned to databases and never actively examined.


This article discusses two approaches to gain more value from all of the data generated in the course of drug discovery projects. The first, multi-parameter optimisation (MPO), enables all of the available compound data to be given appropriate weight in the decision-making process, helping to focus on compounds with the best balance of properties and the highest chance of downstream success [7]. The article will also discuss how compound data, even for compounds that do not progress, can be used to gain information on relationships between compounds’ chemical structures and their properties (known as structure-activity relationships or SARs). These SARs can be used to predict properties of new compounds before synthesis, guiding the design of improved compounds. This allows synthetic and experimental efforts to be focussed on chemistries that are most likely to achieve the required properties.


Multi-Parameter Optimisation


A high quality lead or candidate compound must meet a profile of property criteria including potency against the intended therapeutic target(s), selectivity against potential off-targets, appropriate physicochemical and ADME properties and an absence of toxicity. Unfortunately, these requirements often conflict and finding a perfect compound may be impossible. In addition, the data generated in drug discovery often have significant uncertainty, due to experimental variability or statistical error. Therefore, drug discovery is, in essence, an exercise in achieving a delicate balancing act in the presence of uncertain information.


MPO methods have been developed across many disciplines to address the challenge of simultaneously optimising multiple characteristics and are becoming more widely adopted in drug discovery [8]. An MPO method for drug discovery needs to meet a number of requirements: it must be flexible, because the property criteria for a successful drug will vary widely depending on the therapeutic objective, intended route of administration and many other factors; it must be possible to weight the individual criteria to define acceptable trade-offs


Figure 1. This graph shows the compounds in a data set, ordered by their overall scores against a profile of property criteria shown inset. The score for each compound, representing the likelihood of success against the profile, is plotted on the y axis. Error bars indicate the uncertainty in the overall score for each compound due to the uncertainty in the underlying data. In this case it is possible to see that the error bar for the top-scoring compound overlaps with ~15 compounds in the data set, indicating that these cannot be confidently distinguished. The inset histogram shows the contribution of each property to the overall score for a single compound. In the example shown, the most significant issue to address in order to increase the overall score is solubility (the colours of the histogram bars correspond to the key in the profile).


Structure Activity Relationships


So far this article has discussed an approach to use compound data to select the ‘best’ compounds for progression. However, understanding the SARs that connect compounds’ chemical structures with their measured properties can inform the design and selection of new compounds that have not yet been made.


because, as discussed above, it may be impossible to satisfy all of the criteria simultaneously; the output should be interpretable to provide guidance to scientists regarding the most critical issues to address in order to further optimise a compound; and it must deal appropriately with uncertainty in the data. The last of these is the largest difference between application of MPO to drug discovery and other fields, where the data typically have lower variability and uncertainty. The goal is to identify the compounds with the best chance of success against the over profile of criteria, but without missing opportunities by inappropriately rejecting compounds based on uncertain data.


One approach to MPO in drug discovery is Probabilistic Scoring [7], which enables a project team to define the profile of property criteria that an ideal compound would achieve. These criteria can be individually weighted to reflect the impact of failing to achieve the ideal outcome on the overall chance of success. An example of such a profile is shown in Figure 1. The compound data can then be assessed against this overall profile, taking into account uncertainties and any missing data points, to score the compounds according their likelihood of achieving the project’s objective. Furthermore, the uncertainty in the overall score for each compound can be assessed to clearly identify when the data allow compounds to be clearly distinguished or, alternatively, if more data are required to make a confident decision. One visualisation of these results is also illustrated in Figure 1.


The impact of each individual property on a compound’s score can also be identified, highlighting the most critical issues that should be addressed in order to improve a compound’s overall chance of success. One approach to visualise this information is illustrated by the histogram in Figure 1.


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52  |  Page 53  |  Page 54  |  Page 55  |  Page 56  |  Page 57  |  Page 58  |  Page 59  |  Page 60  |  Page 61  |  Page 62  |  Page 63  |  Page 64  |  Page 65  |  Page 66  |  Page 67  |  Page 68