Poster Abstract

October 16-19, 2005
ACS Prospectives – Advances in Structure-Based Drug Discovery

An interactive evolutionary design environment for lead optimization

James H Wikel, Coalesix Inc., Cambridge, MA

The challenge of developing useful drugs from active molecules is a daunting problem. It is a complex task requiring the simultaneous satisfaction of many criteria including: inherent potency, low side effect profile, good solubility, oral absorption, appropriate distribution in various tissues including the target tissue, metabolism at a rate that ensures sufficient residence time in vivo to be an effective drug, and excretion of parent drug and metabolites in order to minimize any toxic accumulation. The mantra of ‘fail fast’ is especially important in the pharmaceutical industry where the average new drug development costs are currently estimated in excess of $800 million. Associated with the high development costs are also the high risks of clinical failures. For example, an estimated 10,000 new structures are made and tested for every new drug reaching the market. The current success rate for clinical candidates to yield a marketed drug is only 1 in 10 or, put another way, the failure rate for clinical candidates is 90%. Boston Consulting Group estimates that if pharmaceutical companies could effect a 10% reduction in clinical failures they could realize a $100 million savings in development costs for every new drug. The most obvious way to achieve this result is to enter clinical trials with the best possible candidates, whose drug-like properties have been rigorously optimized, thereby improving the chances of a successful outcome. CoalesiX has developed a novel drug candidate optimization technology focused on providing an environment that enables the identification of a diverse selection of preclinical drug candidate molecules with better overall properties, thus supporting better decision making and providing a higher likelihood of clinical success.

CoalesiX's Candidate Design Environment is based on Interactive Evolutionary Computing (IEC). IEC is an optimization technique, which uses subjective human evaluation in problems where it is difficult or impossible to design a fitness function for automatic search of the input space. Evolutionary Computing (EC) employs computer-based problem solving methods that use some of the mechanisms of "evolution" as key
elements in their design. EC systems maintain a population of structures that evolve according to rules of selection and other operators. Each individual in the population has a measure of its fitness in the environment computed through an objective function. IEC adds the additional element in the fitness function that is expressed through the interaction with an experienced discovery scientist.

Drug candidate design and optimization is complex conundrum with two key characteristics, both of which are naturally addressed by the strengths of IEC. First, as described above, a successful drug candidate must satisfy a battery of complex and interrelated in vivo potency and ADMET requirements that are difficult, if not impossible, to describe in a simple computer based model fitness function. While a skilled medicinal chemist can navigate these requirements, progress is slow and inefficient. Second, there are several components of the problem - such as synthetic tractability or IP status – that are subjective or inherently implicit and so do not lend themselves to a purely computational approach.

CoalesiX’s Candidate Design Environment uses IEC to leverage both the expertise of the medicinal chemist and the computational skills of the computational chemist as key elements of this evolutionary search process. The CoalesiX Candidate Design Environment provides for the mutual reinforcement between a) the capabilities of computational technologies to handle large amounts of data and processing (computational scientists) and b) the human ability of skilled medicinal chemists to make judgment calls based upon intuition and experience. The environment creates a comfortable forum for the blending of the computational science inherent in computer based models, as well as the intuition and experience of the medicinal chemists. This talk will describe the architecture of the CoalesiX environment, how the environment fosters productive interaction between the computational and medicinal chemists, and the realization of the environment to enable more rapid and efficient drug discovery and development.

For the medicinal chemist, the technology provides them with the capacity to utilize the input from computational models in an intuitive interface. This results in an interactively guided search through the space of possible structures by a) setting priorities across multiple objectives, b) selecting appropriate predictive tools for various properties and c) providing subjective feedback on structures that the platform generates and presents to him. The Candidate Design Environment is designed to allow for searches of the vast chemistry space defined by the medicinal chemist without exhaustive enumeration of the structures. The Candidate Design Environment follows the directions given by the chemist in terms of tools and priorities and then incorporates the chemist’s subjective feedback into an algorithm-driven automated exploration of the space of potential solutions. It periodically reports back to the chemist the most promising structures that it has found, affording the chemist further opportunity to provide input. The results of this interaction may re-orient the search by changing priorities or by changing parts of the core structure that can be manipulated by the platform. Alternatively, the feedback may reinforce the evolutionary direction for continued structure generation.

In addition, the computational scientists are supplied with experimental results that are to be used to refine and to reposition their algorithms resulting in better, more precise in silico models. New models are then incorporated into the environment, thus the environment is always providing the most current models for use by the medicinal chemistry group. This iterative cycle ensures that all the organizations resources (computational power & experience and intuition) work together, and therefore the candidate design process is fully optimized.

Moreover, the CoalesiX Candidate Design Environment is friendly to IT resources. The environment is built using standard state-of-the-art IT tools. It is designed to operate with minimal support from the IT group and is flexible to accommodate most any standard computer environment. The CoalesiX Candidate Design Environment is not traditional software. It is a cooperative and integrated environment created to provide a synergic focus involving critical R&D resources.

The net result is a move away from the current optimization approach – where a highly iterative and linear process is used, guided and constrained by the anchoring of the medicinal chemist – to that of a design philosophy, where the chemist crafts a solution, prompted by other influences aimed at satisfying all imposed criteria.