In Silico Augmentation of the Drug Development Pipeline: Examples from the study of Acute Inflammation

Abstract

The clinical translation of promising basic biomedical findings, whether derived from reductionist studies in academic laboratories or as the product of extensive high-throughput and -content screens in the biotechnology and pharmaceutical industries, has reached a period of stagnation in which ever higher research and development costs are yielding ever fewer new drugs. Systems biology and computational modeling have been touted as potential avenues by which to break through this logjam. However, few mechanistic computational approaches are utilized in a manner that is fully cognizant of the inherent clinical realities in which the drugs developed through this ostensibly rational process will be ultimately used. In this article, we present a Translational Systems Biology approach to inflammation. This approach is based on the use of mechanistic computational modeling centered on inherent clinical applicability, namely that a unified suite of models can be applied to generate in silico clinical trials, individualized computational models as tools for personalized medicine, and rational drug and device design based on disease mechanism.

Figure 1. Integration of data-driven and mechanistic modeling approaches to modeling disease

Data-driven models may suggest principal drivers or dynamic networks involved in a given disease process, but must be linked via literature data or experiments to create mechanistic, dynamic computational simulations. These simulations may take different forms depending on the process and biological scale (sub-cellular, cellular, tissue, organ, whole-animal) being modeled.

Figure 2. Current and proposed drug development process

Panel A: the current drug development process is both serial and linear, progressing from high-throughput and high-content screening in vitro to testing of candidate compounds in animals and ultimately in clinical trials. Panel B: the proposed drug development process would be parallel and non-linear. It is envisioned that this process could start either from community consensus-based mechanistic computational models of disease, or from mechanistic computational models generated from a combination of high-throughput and high-content data and subsequent data-driven analyses such as Principal Component Analysis or Dynamic Network Analysis (see Fig. 1). In silico clinical trials based on such mechanistic computational models would then be the central component of a parallel process, in which predictions of drug properties and effects would be tested in pre-clinical animal models and subsequently in small clinical studies and ultimately in randomized, placebo-controlled clinical trials. Importantly, the pre-clinical animal models would serve not as surrogates for the human disease, but rather as tests of the validity of the computational models underlying this new drug development process. While existing drug candidates could be screened in this fashion, ultimately the computational models could be used to define the properties of ideal drugs or devices for a given disease or patient sub-group, and the trials of such therapies could be based on individual-specific computational models.

Figure 2. Current and proposed drug development process

Panel A: the current drug development process is both serial and linear, progressing from high-throughput and high-content screening in vitro to testing of candidate compounds in animals and ultimately in clinical trials. Panel B: the proposed drug development process would be parallel and non-linear. It is envisioned that this process could start either from community consensus-based mechanistic computational models of disease, or from mechanistic computational models generated from a combination of high-throughput and high-content data and subsequent data-driven analyses such as Principal Component Analysis or Dynamic Network Analysis (see Fig. 1). In silico clinical trials based on such mechanistic computational models would then be the central component of a parallel process, in which predictions of drug properties and effects would be tested in pre-clinical animal models and subsequently in small clinical studies and ultimately in randomized, placebo-controlled clinical trials. Importantly, the pre-clinical animal models would serve not as surrogates for the human disease, but rather as tests of the validity of the computational models underlying this new drug development process. While existing drug candidates could be screened in this fashion, ultimately the computational models could be used to define the properties of ideal drugs or devices for a given disease or patient sub-group, and the trials of such therapies could be based on individual-specific computational models.

Figure 3. Discovery and development in a high-throughput world

The traditional Scientific Cycle is challenged by the demands of very large, high-throughput/high-content data-sets, and multi-dimensional, complex systems. The technological augmentation of various steps in this cycle requires different methods and operational strategies. Data acquisition and correlative analysis have undergone significant advances in the past few decades. Now, emphasis must be placed on the technological enhancement of hypothesis testing and an engineering approach to therapy development. A data-rich, high-throughput environment calls for a similarly high-throughput approach to hypothesis evaluation and testing: a distributed, parallelized approach based on the application of Translational Systems Biology offers a potential solution. By taking advantage of the evolutionary principles of diversity and selection through culling, a parallelized implementation of the Translational Systems Biology research paradigm can provide a robust and sustainable means of meta-engineering the discovery and development process.