Topic: Data Science
Podium Presentation in Room 3 on Wednesday at 11:00 (Chair: Will Slade)
Authors: Hema Ketha, William O. Slade and Russell P. Grant
Introduction: Contemporary LC-MS/MS workflows lack seamless batch data flow. Batch information is often generated on independent automation systems or, by manually hand scanned sample lists. Manual interventions to batch data cause errors from misaligned batch lists, inadvertently switched rows of samples or subjectivity in chromatographic data review.
Objective: Design and implement a hands-free seamless data flow system to eliminate errors arising from manual batch-accessioning and batch build processes.
Methods: Using applications of R programming, we automated several steps in our LC-MS/MS assay workflow. The Hamilton Star, programmed using the Venus Graphic User Interface (GUI) and Hamilton-Standard Language (HSL), captures and parses barcodes for standards, controls, and patient samples. All steps to these samples, including real-time dynamic TADM error-handling for clots, bubbles, and short samples, are translated to a SQL database. The database is then queried to create a flat-file (*.csv, *.txt), which is then moved to a lab-wide share drive. Once on the share drive, batches can be built for our LC-MS platforms (Aria™).
To connect the flat-file created by the Hamilton to the batch-building process using Aria™ OS, the R programming language was used, including the digest, gWidgets, gWidgetscltk, and XML packages. The operator executes the R program, and a dialog box appears where Batch ID can be scanned. After clicking “ok”, the program scans the share drive for the assay type and presence of that batch. If found, the flat file from the Hamilton is pulled into an R data frame, and the user is then asked to scan the tray position in the autosampler (AS) and select the desired liquid chromatographic system (LC). Once the AS and LC have been selected, the program creates the appropriate XML (batch file) using assay-specific instrument methods and in an assay-specific directory, triggering initiation of our cloud-based data review process using Ascent™.
Results: Data from >1100 batches and >80,000 samples were analyzed using this workflow with automated review on Ascent™ from 08/01/19–11/01/19. Average percent flagged samples were 1.7% (N>1100). %CV for quality control samples (N>1100) was as follows: oxcarbazepine metabolite (10MHD) Low:target mean:14. 41μg/mL, calculated mean: 14.29μg/mL, %CV13.4; high: target mean:62.26μg/mL, calculated mean:59.39μg/mL, %CV8.2; Levetiracetam low:target mean: 8.78μg/mL, calculated mean:9.03μg/mL, %CV22.2; high:target mean:34.1μg/mL, calculated mean: 35.44μg/mL, %CV9.76. The data review process was completely streamlined and the new seamless batch data flow eliminated plate positional errors on the AS. We eliminated errors that can arise from batch scrambling from misaligned or inadvertently switched rows of samples.
Conclusion: This presentation will describe step-by-step development of our “R” based automated work-flow and show how errors that can be eliminated in high throughput LC-MS/MS assays.
|Planning to mention or discuss specific products or technology of the company(ies) listed above:||