I’ve seen many articles come across my desk in the last few months and they all seem to focus on high throughput experimentation being THE ASNWER to the pharmaceutical industry’s productivity problems. The latest was from Chemical Processing on Oct 2006.
Process Engineering | R&D takes the fast track | Chemical Processing
Many of these articles focus on automated systems working overnight and lowering labor costs. While I don’t dispute that, I think there needs to be more thought put into what you can learn from this and how it can be used effectively. In some ways, it reminds me of when desktop publishing first came to personal computers and people used dozens of fonts, formatting and colors simply because they were could. After a while, people realized that using these indiscriminately just because you could, did not result in better communication of ideas. The same holds true with automated high throughput experimentation. Just because you can perform thousands of experiments in a week doesn’t mean you should or that you are getting the best answer. Doing 1,000 reactions in two weeks isn’t as productive as 200 carefully chosen and designed experiments done in three days.
Here are my basic points.
- There is a need to work smarter not faster.
- Better use needs to be made of the data generated.
First, I am a firm believer in spending time up front thinking about the problem before starting to do experimentation. This avoids the “Ready, Fire Aim” mentality that seems to be so pervasive in some R&D departments. I understand this tendency and fell into that trap myself early in my career. However, taking the time to think through the possible variables that could have an impact is a worthwhile exercise.
One of the most important aspects of this thinking ahead before performing experiments involves, design of experiments. This is a way of running reactions to get the greatest amount of information from the smallest number of experiments. While it may not be as imperative as it was in the past to run a small number of experiments, it still is valuable to run the best set of experiments possible. Using advanced software, it is possible to get the best set of experiments regardless of how many variables there may be or whether or not those variables are discrete in nature (such as which catalyst to use) or non-continueous (where only a given set of of conditions can be ran). I would point out that QD Information Services has the capability to design experiments specifically for your situation.
The second point involves better use of the data generated. I have come across too many situations where the approach was to run hundred or thousands of experiments and then sort those in a spreadsheet and go forward with the top one or two choices. At a major process development conference last year, one of the top ten pharmaceutical process development directors made a point of listing all the reactions they had run to find the “best results”. At the end I stood up and asked what they did with the data and got the reply that basically they had solved the problem and moved on to the next problem.
The problems with this approach is it does not take advantage of all the experiments that have been performed. Why invest the amount of time into setting up that many experiments and collect that much data and not try to get as much information as possible out of it? Also, how do you know you have found the “best answer”? Too many times, it is settling for what is acceptable instead of what is best. This leads to openings for your competitors to come in, find the true optimum and possibly patenting that and preventing you from using the best possible conditions.
The best approach is to design the experiments up front but even if that doesn’t occur it is possible to do data mining and possibly find meaning in the data. It continues to amaze me the number of large and mid-sized pharmaceutical companies that don’t take advantage of all the data generated to get the most out of their investment. And too many times, the answer I get is they don’t have time to do more and must move on to the next problem. I often offer to do this data mining for them but they still fail to see any value in this activity. To me, I think this is poor decision making and not getting the most out of your investment and it could come back to haunt you.
I just want to remind folks that QD Information Services can help in these sorts of situation so if you have a need for this, please feel free to contact me and I’ll help you understand how you can get the most out of your data.
Technorati Tags: data mining, design of experiments, high throughput
Email Ed Vawter
Related Posts:
Schering Plough To Stay IndependentReasons I Got Into The Pharmaceutical IndustryDoes Revamping Big Pharma Make Them More Efficent?AstraZeneca to purchase Shire? Another Potential MergerLilly Announces Strategic Changes
Bookmark it with:
| 
del.icio.us
| 
Digg it
| 
Furl
| 
Simpy
| 
Spurl
| 
My Yahoo!
|
