Facing an opportunity that would make many informatics managers green with envy — building an IT infrastructure from scratch without the burden of legacy tools or data — Les Klimczak was careful not to take his good fortune for granted. When designing the informatics system for Gaithersburg, Md.-based drug discovery startup Psychiatric Genomics, Klimczak said he took extra steps to ensure that it could grow and adapt as the small company’s needs change.
Klimczak, Psychiatric Genomics’ director of bioinformatics and IT, set out to overcome several primary challenges he identified for startup informatics. At the top of his list was the difficulty in meeting tight timelines with limited resources. The solution to this problem, he noted, is acknowledging that “you can’t integrate everything — only the specific things that impact your discovery platform.” Another key solution to this hurdle, he said, was making the system extensible so that it can accommodate future growth when appropriate.
One shortcut to integration, Klimczak pointed out, is at the experimental design stage, where a well-thought-out experiment can essentially result in pre-integration of the resulting data. Biologists at Psychiatric Genomics design their experiments with the help of analysts, who ensure that the results will have a common denominator so that they can be compared.
Another problem he identified was that advanced analytical tools “are not always accessible or understandable for biologists.” Since user-friendliness and advanced analysis rarely go hand-in-hand, Klimczak said his approach is to hire analysts who are trained in biology to pre-analyze the raw data and then deliver the results to the biologists. Psychiatric Genomics built a user interface that supports a limited number of “generic” analytical scenarios for the biologist end-users, he said, simplifying their view of the pre-processed data.
One of Klimczak’s biggest obstacles was a commercial LIMS from an unidentified manufacturer, which limited the availability of other software tools because the LIMS couldn’t “talk to” the analytical tools. “It was easy to get the data into the LIMS,” he said, “but very difficult to get it out.” Klimczak’s solution was to build a parallel system that shared a user interface with the LIMS, but offered a separate set of data warehouses, data marts, and analytical pipelines.
Klimczak said that he has looked at LIMS offerings from other vendors, but none of them appear to address the drawbacks he has encountered. But in the meantime, he said, the extensible system he built ought to work just fine.
— BT