This project is representative of the dynamic nature of pharma research informatics. Multiple commercial software tools for data capture, processing, and archival are purchased - each for its strengths - and then these tools coexist in a fast-paced lab environment. Small inefficiencies in data processing workflows become significant bottlenecks measured in hours and sometimes FTE days lost every week attending to repeated manual tasks.
Our Client's screening biologists collect response data in a bioassay at multiple concentrations and then analyze it to determine inhibition constants. This of course is done by fitting dose-response curves. For small numbers of experiments the biologists can use desktop software then manually enter the results into the corporate assay registration database. When there are 50 curves a week to fit, many with noisy data that needs careful review, the process starts to get tedious. Eventually the lab is asked to collect results for hundreds of curves per day in a new screening campaign, and at that point manual curve calculation and review becomes impossible. This is exactly what happened in this lab.
Small inefficiencies, when they reoccur constantly, can disrupt the entire workflow.
We worked with the Client's computational scientists to identify bottlenecks in the current data processing workflow. Together we designed and implemented an automated web application to capture assay data from files, automatically fit dose-response data, then present it for review and correction to the biologist in a visual format suitable for hundreds of curves at a time. Most curves are fit correctly by the algorithm and need only be viewed quickly. The difficult cases are color-highlighted on the screen using a heuristic algorithm, and the biologist can correct the fit using one of several predefined combinations of parameters, usually with only one or two clicks. This simple approach increased data reviewer's overall productivity between ten- and twenty-fold depending on the assay.
In this project the key piece to a solution was to identify and understand precisely what was the repeated step that data reviewers had to go through in their workflow. Once that was identified, we changed the software-driven workflow to perform most of the work automatically and expose to the human reviewer the most difficult curves, with pre-packaged actions readily available on-screen.
