Translating between preclinical data and clinical outcomes: The bottleneck is always on the other side

Drug development is an information-rich endeavor, and it isthe role of informatics researchers to ensure we are leveraging importantfindings from internal and external sources. Our work enables better-informeddecisions about compounds in the pipeline, and it's especially beneficial intranslational medicine, helping to improve the likelihood of success ascandidate drugs move into clinical development.

Translational medicine provides a better understanding ofdrug mechanisms and interactions, which aligns well with today's regulatoryemphasis on safe medicines with fewer side effects. By being able to translatepreclinical data and observations into possible clinical outcomes, we can makethe drug development process more efficient and cost-effective. Our workcontributes to the goal of reducing attrition rates, improving our ability topick winners and drop losers, in earlier and less-costly phases of clinical trials.

Translation needs to happen in both directions: Forwardtranslation, or translating from preclinical studies to patient studies, andback-translation or "feedback loop," in which patient data from clinicalstudies are used to "humanize" preclinical drug discovery. The biggestchallenge today is finding or accessing all the relevant and necessary data andinformation, and we often spend more time getting to the data than using it.

I started out in preclinical discovery many years ago, and Iwas surprised at the difficulty in finding out what happens to compounds oncethey enter the clinic. Even though high-level results from studies wereavailable, we couldn't really see all the critical values. That's when Istarted looking into ways to improve information flow. I moved into clinicaldevelopment to understand the problem from the "other side."

What I found wasthat depending on where you sit, the bottleneck is always on the other side.

After spending almost six years in clinical development, Irecently moved back to the preclinical side, with better knowledge andappreciation of where the clinical data is— and the ethical and legal aspectsof patient privacy. My role now includes responsibility for improving theinformation flow between these two sides so both can use the information in thebest way possible. Ultimately, I want to eliminate any bottleneck so thatwhichever side you're sitting on, you have access to information that cananswer your questions, progress projects and deliver new medicines to patients.

The questions raised in translational medicine are key todeveloping the targeted therapies of personalized medicine. Working withselected populations, you might need to identify improved responses andsurvival rates, see what effect a compound has on a particular type of tumor,and gauge performance against currently marketed products.

The ability to find answers relies on data diversity andcomplexity. Identifying preclinical models—in silico, in vitro, in vivo—or assays that can best predict clinicalobservations is not trivial. It requires understanding ofpreclinical-to-clinical correlations at the project level, at the model/assaylevel and at the subject level. The challenge in making these correlations liesin what we call the "translational chasm"—the information gap betweenpreclinical and clinical information.

Maintaining an ongoing dialog betweenpreclinical and clinical scientists, so they can improve the efficacy or safetyprofiles of candidate drugs, is challenging and involves three interrelatedaspects.

The solution has to address the technology difference,cultural difference and skills difference within each organization. We oftentalk about technology and culture, but even the right culture—in terms ofwillingness and understanding—with all the right tools can fail if we don'thave the right skills. Not many scientists have the necessary computational,quantitative and information science skills necessary.

The complexity of the data requires more than "point andclick"—even with the most sophisticated tools—to access, retrieve, integrate,and analyze the necessary information and knowledge. It requires trainedscientists who fully comprehend the scientific or medical problem and also havestrong information and computational skills and expertise to applysophisticated multidimensional analysis and visualization approaches to thedata.

While technology is part of the solution, it has created itsshare of problems. A lack of standards among different systems and programs hasled to technology bottlenecks in the information flow. The good news is we'reseeing some convergence and progression of standards, allowing disparatesystems to communicate. While progress is being made across the board, it hasbeen in pockets, so the overall effect is one of small steps.

Because all of us within the industry are dealing withsimilar issues, several consortia have come together to join efforts and createlarger datasets. When researchers look for correlations, it's like searchingfor a needle in a haystack. With a big enough haystack, you might find severalneedles to compare for similarities. With large enough datasets, researcherscan identify more of the factors that contribute to the success or failure of adrug, or a particular class of drugs, thus gaining better insights.

This pooling of data across the industry might seem likecompetitors are now collaborators, but the sharing is "fit for purpose,"without revealing competitive advantage or proprietary secrets. Thecollaboration is toward a common goal of developing a better understanding ofthe science and disease, not products. Working alone, no company can achievethe richness of data and knowledge that these consortia can achieve together.

As information flow improves between preclinical data andclinical outcomes, the need for even better communications becomes apparent.Today, we have a good handle on where the data reside and they can flow to anyparticular project. The next and more important leap is working at a higherlevel, so project teams can make decisions in the context of everything elsethat may be related. We need to ensure access to data across projects andacross functions, even including inactive projects because there is valuableknowledge in our legacy projects.

The outline of the task is relatively simple, if difficultto achieve. We need to extract and structure our project knowledge in a waythat enables translation. We need to connect project information acrossfunctional and scientific boundaries. We need the skills and ability to mineinformation across boundaries.

Today's knowledge base is a complex system of data sources,abstraction pipelines, document management, taxonomies/ontologies, text-miningengines, curation/quality control tools, and query capabilities. The knowledgebase of the future needs to also integrate the clinical data and knowledge,much of which sits securely in regulatory-compliant document repositories.

To be successful in translational science, we can't waituntil all of the necessary infrastructure is in place. We can build on the hardwork and hand curation of data currently done in translational medicine. Wealso need to recognize that finding correlations in complex data requires largedatasets, not just diverse data, and can be aided by exploiting existingknowledge and legacy data.

The issue of bottlenecks (existing and potential ones) isever-present. The solution comes from having the right technology, culture andskills to address the problem. Most people focus on the technology and theculture. But it is the skill to understand the questions, process the data, andapply computational and quantitative approaches to exploit the data in anintegrated fashion, that allows us to resolve bottlenecks wherever they existand to provide context that makes a difference in the drug development process.

You can have the best tools possible, but if you don't havethe skills to use them—or don't fully understand the question at hand—therewill always be a bottleneck on one side or another.

Dr. Anastasia Christianson is a senior principal scientistin informatics and senior director of discovery information at AstraZenecaPharmaceuticals based in Wilmington, Del. Christianson began her professionalcareer at a small biotech company, DNX Biotherapeutics Inc., in 1992 beforemoving to Zeneca's Pulmonary Pharmacology department in 1994. Since then, shehas helped to set up AZ's Genomics and Bioinformatics division in Wilmington,helped to establish informatics expertise in experimental medicine and clinicaldevelopment, and has been the biomedical informatics global discipline leaderfor the last five years. Outside AstraZeneca, Anastasia has held adjunctprofessor appointments at universities including Johns Hopkins University,University of Pennsylvania and Drexel University. Christianson obtained herPh.D. in Biological Chemistry from the University of Pennsylvania in 1989,followed by postdoctoral training at Harvard University in Cellular andDevelopmental Biology.