Validation of an automated gating pipeline for ICS assays in vaccine trials
Year: 2017
Type: other
Abstract: High dimensional flow data is becoming ubiquitous, and many new computational methods have been developed to mine, visualize and make sense of it. In clinical and pre-clinical trials however, manual gating remains the gold standard. The focus in these studies is accurate, precise, robust, and reproducible identification of pre-defined biomarkers to be evaluated as trial endpoints. The majority of automated gating methods are focused on discovery of new cell populations and the trials-centric segment of flow cytometry has been under-served by the computational community. For the past few years we have been developing OpenCyto, a collection of robust, open-source software components that function as a framework for data management, analysis, and automated, data-driven identification of pre-specified cell populations. We have demonstrated that OpenCyto is reproducible, accurate and robust, and can improve upon the performance of manual gating. While still relying on manual gating, we have leveraged the data management components of OpenCyto for reproducible data analysis within the HVTN (HIV Vaccine Trials Network) and the (CAVD) Collaboration for AIDS Vaccine Discovery. OpenCyto uses a domain-specific language to define "gating templates", which are interpreted to perform data-driven gating of cyotmetry data. Defining these templates requires translating expert knowledge across domains; the user needs to select gating algorithms and their tuning parameters, an impractical task when constructing a complex analysis scheme. However, manually gated data is an rich source of expert knowledge, and now we have developed tools to define these templates and learn their parameters from manually gated data. We have applied these tools to a high-dimensional ICS assay from a study evaluating vaccine response. We show that training on as few as 15 samples we are able to recapitulate over 165 manually gated populations in over 380 test samples, and that discrepancies have little impact on downstream inference, demonstrating the robustness of the approach. This pipeline leverages the flow cytometry infrastructure in Bioconductor and demonstrates how these tools are used to address real-world problems. This work represents an important step in the automation of flow cytometry data analytics.
Institution: Fred Hutch Cancer Center
Language: English
Subfield: Biophysics
Domain: Life Sciences
Open Access status: gold
OpenAlex ID: W2736091601