09 Jan 2006
As I commented in an earlier post, we’ve been transitioning our legacy EJB-based system to a new platform built largely on Spring and leverging hibernate for persistence.
The technology has worked well, I’ve seen improvements by way of code reduction and ease of implementation in certain areas. Instead of 500 remote methods exposed via 30-odd session beans (and stateful ones at that for a reason I won’t get into now), I’ve implemented a fairly straight forward command pattern variation that combines a metadata layer with EJB/MDB and Spring execution end-points. Persistence is easier, and a custom DSL built-on top of HQL has allowed us to replace nearly all of our EJBQL and CMP-code generically.
However, all is not bliss. We’re suffering from a bit of technical debt. We attempted somewhat of a model-driven approach with effort put into developing a UML model of our existing schema (and some sequence/class diagrams). The original goal was to use andromda tags to annotate our model and produce useful hibernate mapping files. It sounded good in practice, but of course it didn’t work as expected (and really wasn’t that big of a surprise either). We’ve annotated a lot of the model, but there are a number of hibernate features not supported by andromda that required some post-processing tweaking in Perl.
The other significant drawback I’ve noticed is slowness at initialization time. I know that we’ve introduced a lot of overhead in the parsing and interpretation of our own metadata layer, but there appears to be sufficient overhead in the initialization and preparation of the 130-odd Hibernate mapping files and 40-50 Spring beans. Disabling cglib has not shown much in the way of improvement, but we’re still investigating and profiling.
The last problem is one that was faced even in the old legacy system. Every 5 or 6 deploys we get an out of memory exception in JBoss. In the legacy system it was a regular OutOfMemoryException, in the new hibernate/spring platform (still deployed to JBoss with an EJB3 SLSB and MDB) we get out of memory problems in the Permgen space. I’m not the only one with this problem and it’s affecting all of our developers. I’ve bumped up the memory allocated to permgen to 128MB but that only delays the problem.
If anyone has any suggestions or ideas, feel free to comment.
19 Dec 2005
My buddy Glenn is actively travelling in India right now so I thought I’d post this Wired article about Pharmaceutical’s using India as a clinical test bed. We both work in Bio-IT and although we’re not directly involved in drug discovery, we do develop tools to help track and analyze researcher’s data.
Testing Drugs on India’s Poor
I wonder how much of it is true and just what can be done to stop it. In today’s global marketplace and given the emergence of India as an educated yet poverty-stricken nation, I don’t see why large pharma wouldn’t take the 60% cost savings and do a trial in India. Clinical trials are expensive and given that only a small percentage of them are successful, you’re forced to be operate efficiently.
The article goes on to mention that clincial trial exports to India are expected to top $2 billion by 2010.
In 2004, two India-based pharmaceutical companies, Shantha Biotech in Hyderabad and Biocon in Bangalore, came under scrutiny for conducting illegal clinical trials that led to eight deaths.
In another incident, Sun Pharmaceuticals convinced doctors to prescribe Letrozole, a breast cancer drug, to more than 400 women as a fertility treatment in a covert clinical trial — and used the results to promote the drug for the unapproved use.
Neither of those stories sound very good. You’ve got pharmaceutical companies inticing relatively poor and “treatment naïve,” people. Not a very good recipe for discovery and clinical testing of drugs that are going to be sold and distributed in the global marketplace.
11 Dec 2005
People ask me all the time what I do.
I essentially tell them I’m a technical lead at a Bio IT software company that develops solutions for lab management. If the conversation goes any further, I’ll mention that we’re specifically targeting proteomics facilities and their significant data management and process tracking needs.
I was just doing some reading tonight and came across a pretty good description of proteomics. I’ll post it here for the interest of others: (the content actually came from a post about the Dana-Farber Institute)
Though not a new field of study, proteomics is poised to have a potentially sweeping effect on cancer research, thanks to recent advances in both technology and the ability to analyze mammoth amounts of data. In recent years, Dana-Farber has made a firm commitment to the field with the purchase of equipment capable of rapidly analyzing the proteins made by cells, and with support of research that aims to identify some of the hallmark proteins associated with cancer. As part of this effort, the Institute recruited Jarrod Marto, PhD, and John Quackenbush, PhD, experts in proteomics and computational biology (which develops computer models to analyze large data sets), respectively. Marto will direct the new center.
While genomics focuses on the activity of the approximately 25,000 genes in human cells, proteomics is concerned with proteins, the “workhorses” of cell life, which carry out a cell’s functions, be it transporting oxygen (as in red blood cells), filtering toxins from blood (in liver cells), or secreting digestive acid (in stomach cells).
Genes issue the instructions for protein production, but genomics provides little information on when or how much of a protein is made. Moreover, proteins’ function depends on their interactions with other proteins and on modifications they undergo over time — areas about which genomics is silent. Finally, most drugs, including anticancer drugs, are directed against proteins, so a more complete understanding of the proteins that contribute to the behavior of cancers will result in better and more effective drugs.
Proteomics seeks to identify all proteins produced by cells, as well as their interacting partners and structural changes, as these change over a cell’s life cycle and as normal cells become cancerous. Such information may prove invaluable in designing new cancer treatments and diagnostic tests.