Biotechnology from bench to business

• Follow GEN on:
• Login | Register | Subscribe

Improving R&D with Better Decision Making

Helping Scientists to Overcome Their Human Cognitive Biases

• Matthew Segall
• ,
• Edmund Champness

(Page 2 of 3)

Guiding Decisions

Figure 2. A “scoring profile” showing the properties of interest, the project’s success criteria, and the importance of each to the project’s objective: The inset window shows how more subtle trade-offs than simple pass/fail criteria can be defined, in this case a range of values over which the property value goes from ideal to unacceptable. Click Image To Enlarge +

Figure 2. A “scoring profile” showing the properties of interest, the project’s success criteria, and the importance of each to the project’s objective: The inset window shows how more subtle trade-offs than simple pass/fail criteria can be defined, in this case a range of values over which the property value goes from ideal to unacceptable.

The process of guiding decisions, in contrast to supporting decisions, begins with a definition of the objectives of a project, defined as the property criteria that the project team would ideally like to achieve. Interactive software, such as StarDrop, can use this definition to proactively guide the decision maker to focus effort on the options, in this case compounds, which are most likely to achieve the required balance of properties.

A sample “scoring profile” shown in Figure 2, defines the properties of interest and the success criterion for each. In addition to the criteria, their relative importance to the success of the project is also defined, as in practice it is often necessary to make a trade-off between properties if an ideal molecule cannot be identified. The available data can then be combined into a single score that reflects the overall quality of a compound against this profile.

It is also essential to take into account the uncertainty in the underlying data. When all of the data is combined in a single score, scientists need to consider the resolution this provides to distinguish between compounds. To achieve this, an overall uncertainty in the score for each compound can be calculated.

Figure 3. A screenshot of StarDrop showing the output of scoring the compound set using the profile in Figure 2. Three example visualizations are shown. The graph on the right shows the scores for all 267 compounds along with error bars indicating the overall uncertainty in the score. From this, the highest quality compounds can be clearly identified and it can be seen that the top ~25 compounds (highlighted in green) cannot be confidently distinguished from the top compound based on the available data. The chemical space on the left shows the diversity of the chemistry in this set, colored by the score from highest (yellow) to lowest (red). This allows areas of chemistry with a high chance of success to be clearly identified. Finally, the histograms for each compound (example shown enlarged) highlight key issues to overcome in order to significantly improve the quality of the compound. In the highlighted example the light blue and pink bars are lowest, indicating issues with logP and hERG inhibition (see key in Figure 2). Click Image To Enlarge +

Figure 3. A screenshot of StarDrop showing the output of scoring the compound set using the profile in Figure 2. Three example visualizations are shown. The graph on the right shows the scores for all 267 compounds along with error bars indicating the overall uncertainty in the score. From this, the highest quality compounds can be clearly identified and it can be seen that the top ~25 compounds (highlighted in green) cannot be confidently distinguished from the top compound based on the available data. The chemical space on the left shows the diversity of the chemistry in this set, colored by the score from highest (yellow) to lowest (red). This allows areas of chemistry with a high chance of success to be clearly identified. Finally, the histograms for each compound (example shown enlarged) highlight key issues to overcome in order to significantly improve the quality of the compound. In the highlighted example the light blue and pink bars are lowest, indicating issues with logP and hERG inhibition (see key in Figure 2).

Of course, it is essential to support this analysis with visualizations that help scientists to identify patterns in their chemistry and identify compounds on which to focus their attention. One of these is shown in the graph in Figure 3, in which the scores for each compound are plotted, along with error bars that show the uncertainty in each score indicating the confidence with which compounds can be distinguished.

Plotting this information in a chemical space, which reflects the diversity of the chemistry being explored, allows hot spots to be quickly identified in which high-quality compounds are most likely to be found. Furthermore, the impact of each property on the individual compound scores can also be identified, taking into account not only the property value, but also its uncertainty and the importance of the property. This highlights which properties would have the largest impact on the overall quality of a compound if improved. Examples of these visualizations are shown in a screenshot of Optibrium’s StarDrop software (Figure 3).