Artifical Intelligence In Drug Design

Whilst there is a strong tacit dimension to the medicinal chemist's knowledge, many of the tactics and techniques used to navigate chemical space that derive from experience can be articulated and therefore potentially formalized. If we can articulate the tactics used to explore SARs efficiently then the next step is to formalize and codify such knowledge. Codifying medicinal chemistry knowledge enables the goal of evolving from computer-aided drug design (CADD) to knowledge-aided drug design (KADD).

Methods for developing KADD may be found in the field of artificial intelligence. Two features of medicinal chemistry hope that artificial intelligence methods could aide the drug designer. First, drug design is often a problem with a complete description and second the rigorous scientific method that medicinal chemists apply in during compound optimization. The computer scientist's definition of a problem with a complete description is one for which we can define the specification of the solution [69]. The medicinal chemistry solution - a desired drug candidate profile -can be accurately described as, for example, ''an oral, rule-of-five compliant, inhibitor of phosphodiesterase 5 with 100-fold selectivity over phosphodiesterases and with no structural alerts present". Even though the problem can be completely described, the difficult challenge is to find an optimal solution. The difficulty of this class of problem, with complete descriptions, is one of effectively navigating the vast space of feasible solutions (referred to as a nondeterministic polynomial (NP) hard problem) [69].

Recent advances in artificial intelligence have demonstrated that parts of the iterative scientific discovery process can itself be the subject of automation. In particular, the "closed loop'' of hypothesis generation and in vitro experimentation in drug design is potentially amenable to improvement through knowledge-based artificial intelligence. The automation of the iterative hypothetico-deductive process [70], using advances in inductive logic programming [71], has recently been demonstrated in the field of functional genomics [72]. A "robot scientist'' was designed to elucidate cellular metabolic pathways. The system applied inductive logic programming to learn from a large knowledge base of metabolic pathway information in order to propose hypotheses of how the pathway could be constructed. The system tested the hypothesis by ordering reagent and running the experiments through an automated-laboratory robot. Results from the experiments were fed back into the hypothesis generation engine, to design the next round of experiments. Reagents costs could also be factored into the experimental design enabling the system to design a set of experiments that could either get to the answer by the quickest or cheapest route. There have also been theoretical developments in attempts to formalize the creative process in the field of computer-aided design in engineering [73]. These developments are intriguing as the hypothetico-deductive process of medicinal chemistry drug design is most analogous to the functional genomics problem solved by the robot scientist. Theoretically, what would be required to develop an intelligent interactive design tool for medicinal chemists?

In all creative fields the design process decreases the possible number of solutions between the current state of the product and its desired specifications. The medicinal chemist intuitively understands that identifying and selecting the "lead" is the crucial in step increasing the likelihood of the future success of a project. Selection of the right chemical lead reduces the search space enormously. The history of medicinal chemistry is replete with examples of one or two subsistent changes between the lead and the final drug product [74,75]. If we consider drug design as a multi-dimensional optimization problem then combinations of probabilistic predictions and heuristic search patterns (e.g. evolutionary and emergent algorithms) can be employed to explore pathways through "hypothesis space''.

Inductive logic programming [71,72] is a suitable method of hypothesis generation that is capable of learning from large-scale knowledge bases. However, a key problem is what has been called the "knowledge-acquisition bottleneck". In order to reduce the search of "hypothesis space'' medicinal chemists, utilize knowledge of "tactics", which can be applied in many situations, to make large conceptual jumps in the search space. Common tactics employed by the experienced medicinal chemist include ''methylene shuffle'', adding lipophilicity, adding chirality, searching for hydrogen-bond interactions, introducing or breaking conformational constraints, amongst many others [76]. The advent of large-scale semantically normalized integrated databases [17] provides a resource for such automatic knowledge acquisition and data mining for the discovery of new tactics and rules. In order for a design system to learn and apply the tactics of the medicinal chemist, a suitable

''chemical algebra'' to describe transformations in drug design would need to be derived [77]. Such a ''chemical algebra'' could be found in semiotics (the science of signs and processing of signs) which being explored, for example, in engineering to establishing a computational framework for design theory [73]. The advantage of applying semiotics is that algebraic semiotics [73,78] could be a suitable tool to manipulate such a ''chemical algebra''. In order to encode medicinal chemistry knowledge as semiotics a meta-level representation of chemical structure - what could be called ''chemiotics'' - is required. The costs of the iterative rounds of experiment can also be judged by connecting the hypothesis generation engine to various established retro-synthesis programs or virtual chemical libraries software that can estimate the synthetic accessibility.

Of the theoretical requirements for building an intelligent KADD system, as outlined above, the medicinal chemistry knowledge bases, the evolutionary search engines and the hypothesis generation methods are in existence today. What currently missing is the semiotic representation of medicinal chemistry knowledge in terms of tactics and rules. The application of knowledge-based artificial intelligence to medicinal chemistry design would enable the insights and knowledge of expert drug designers to be leveraged across a larger number of projects. Alternatively, less-experienced drug designers could be guided in their design by systems build on a wealth of historical knowledge, combined with predictive modeling [29,79]. By formalizing medicinal chemistry knowledge in an intelligent system, a radical improvement in productivity should be possible.

SEO Skills And Mastery

SEO Skills And Mastery

Get More Traffic With SEO. SEO is short for search engine optimization. This is the very complex yet very visible way of accessing websites or web pages within the natural or unpaid realm of search results. Simply put in its literal sense, is that the more visits or hits as it is often referred to the more visible the said site would be when a search is applied.

Get My Free Ebook


Post a comment