Integrating Everything: The Molecule Selection Toolkit, a System for Compound Prioritization in Drug Discovery

Abstract

In recent years there have been numerous papers on the topic of multiattribute optimization in pharmaceutical discovery chemistry, applied to compound prioritization. Many solutions proposed are static in nature; fixed functions are proposed for general purpose use. As needs change, these are modified and proposed as the latest enhancement. Rather than producing one more set of static functions, this work proposes a flexible approach to prioritizing compounds. Most published approaches also lack a design component. This work describes a comprehensive implementation that includes predictive modeling, multiattribute optimization, and modern statistical design. This gives a complete package for effectively prioritizing compounds for lead generation and lead optimization. The approach described has been used at our company in various stages of discovery since 2001. An adaptable system alleviates the need for different static solutions, each of which inevitably must be updated as the needs of a project change.

From flamingo dance to (desirable) drug discovery: a nature-inspired approach

“…Despite their applicability, the Lipinski Ro5 as well as other related approaches reflect a static view of drug discovery [11] that is based on a compound-centric perspective typical of lead optimization projects [14]. However, to quantify the progression of lead optimization projects through process-centric analysis, statistical frameworks are needed.…”

“…It transforms an objective function to a scale-free desirability value, which measures the decision-maker satisfaction against the objective value [11]. In the context of drug design, the decision maker is the chemist and the objective functions, as shown in Table 1, refer to the endpoint values, which can be experimentally measured or theoretically predicted.…”

“…In contrast to the binary pass/fail outcome of hard filters, this approach provides a continuous desirability scale accounting for even slight changes in the value of the endpoint. This enables in-depth and more informative compound analysis and quality assessment [11]. …”

“…A score in the interval [0,1] is given to all the other compounds completing the piecewise linear desirability function. Irrespective of whether the piecewise desirability function is linear, sigmoidal, or of another form, the use of the desirability function has the benefit of smoothing endpoint values compared with to hard filters [11]. Once each endpoint has been assigned a [0, 1] desirability score, all the endpoints can be combined into an overall weighted desirability.…”

“…Many recent developments have focused on methods to aid the simultaneous optimization of multiple factors required in a successful drug, targeting compounds with the highest chance of downstream success early during the discovery process [1]. However, formalized MCO approaches are not widely used in drug design [11]. Thus, here provide the ‘anatomy’ and potential scope of methods for MCO in drug discovery.…”