A recent article in J Med Chem by A. Maynard at GSK describes a statistical framework which can be used to quantify and visualise through process-centric analysis the progression of lead optimisation (LO) projects. A. Maynard and his team at GSK propose a framework to visualise the whole lead optimisation process using compounds’ activities, physical properties, DMPK and pharmacology risk alongside ‘design entropy’ describing it as the ‘LO telemetry’ of the project. This is a shift from looking at a lead optimisation program in pure static compound-centric terms such as Lipinski’s Rule of 5. In a LO project each subsequent compound synthesised is an end-point in a succession of many compounds and by analysing the dynamics of project progression allows better decisions to be made regarding actual efficiency and potential of the project.
Aiding managerial decisions on LO projects
An article in Nature Reviews Drug Discovery by R. Peck in 2015 discussed that behavioural, cultural and organizational issues in industrial research were key obstacles in preventing termination of projects at earlier points. An argument made by Peck on why it is difficult to terminate failing projects is due to the subjective biases even scientists are subject to. For example, optimism causes overestimation of the probability and timeline of success.
The motivation of Maynard to develop a quantitative view of LO progression can aid in more objective decisions when evaluating LO projects. There is certainly a need to develop more process-centred evaluation tools that enhance efficiency and productivity in drug discovery. This is not an argument for increasing management where efforts are only being made to ‘game’ the metrics system in order to achieve targets. Instead, by using previous projects to understand progression, failure and success, metric tools can be developed to aid managerial decisions in future projects.
Quantify, visualising and monitoring LO projects
LO aims for convergence to its endpoints. In Maynard’s paper, LO convergence was quantified through the statistical minimisation of risk. So for a given optimisation variable (e.g. herg_pEC50), there is a distribution of SAR, some of which is closer to the end point. As the optimisation process moves on, the SAR moves closer to the end point eventually achieving convergence. The ‘risk’ of a given optimisation variable is its resistance to convergence which is quantified by how the mean of the SAR distribution moves closer to the end point. Ultimately, the aim is for projects to converge to a score of 0 otherwise failure to reach an endpoint for a particular endpoint means a residual risk is carried. Essentially, all that is needed are variables and their associated convergence end points. This consist of in vitro and in vivo DMPK variables generic to most projects and then more unique target variables such as potency and off target selectivity. Furthermore, physical properties (e.g. solubility) can also be included.
The paper describes four LO programs targeting hepatitis C virus (HCV) replication inhibitors. Protein targets were NS4B, NS5A, NS5B and PI4Ka. NS5A and NS5B were successful while NS4B and PI4Ka where halted due to preclinical safety. An additional oncology project, where the target was not named, was also followed. This project failed due to LO tractability. The paper gives further information on each of these projects and follows their LO telemetry. For example, Figure 1 shows the evolution of the NS4B project towards convergence. Each step in the staircase indicates the next lead compound and then this is followed by more SAR with close analogues until the next lead is found.
Figure 2 illustrates this same picture through the multiple risks involved and show the interplay between them.
Another metric used, ‘design entropy’ is based on the principal that diversity of chemical space explored can relate to convergence. Projects will start off exploration around SAR where there is a higher chemical diversity in structures explored. Eventually, a local minimum is reached and exploration becomes very conservative as a project zeroes in on convergence to the end point. But if a team becomes stuck on a given variable, diversity of chemical matter will increase again to escape this issue until they find an alternative local minimum where again more conservative changes will be explored. Here, chemical diversity was calculated in terms of Shannon’s entropy where 1024 bit chemical functional group fingerprints was used to compute entropy.
In the NS4B project, the design entropy reached a minimal near compound 4, but at this point, toxicity was discovered and so design entropy increased as attempts were made to escape this series.
Currently, progression is measured by milestones that fall in to ‘static’ compound endpoints. The idea here is for the progression of a project to be readily visualised and tracked which can have the benefits of supporting portfolio management, provide support to individual projects and help co-ordinate LO programs. More data on successful and failing projects can allow for more sophisticated analytical tools in order to reduce the amount of time spent projects that will fail.
There is no doubt that these sorts of managerial tools are used with caution and in context and not used as the main determinant in assessing performance of chemists. Experience has taught us that this can drive a dopamine driven mentality of chasing targets much like the banker’s behaviours in 90s and early millennium which led to the financial crash is 2008.
These sort of metrics once further developed can potentially be adjusted to be suitable academic drug discovery, where projects tend to involve the development of much less chemical matter due to limited resources. This could lead to improved productivity through the use of these metrics and a bigger focus on management which is often less seen in academia.
Any opinions noted are those of the blog author only.
Blog written by Yusuf Ali