Will automation knock off manual evaluation?

Abstract

The manual evaluation of hundreds of polychromatic samples for immune cell phenotyping and analysis of data from high-throughput high-content imaging is not only time consuming but also error-prone. Even if manual gating is purely subjective, because the cytometrist begins the evaluation with a certain expectation, it is still regarded as the gold standard in cytometry. From the technical point instrumentation is developing rapidly; every year new cytometers reach the market equipped with even more lasers and more detectors to enable increasingly complex multi-color measurements. Gaps in instrumental requirements are constantly closed, as exemplified by the works of Zhou and coworkers (this issue, page 419) and Rebelo and coworkers (this issue, page 437). But why aren't respective evaluation software tools coming into the work-flow at a comparable speed? Is it due to mistrust of cytometrists to use these software tools? Do trained technicians have difficulty trusting unknown algorithms that will replace their sequential two-dimensional analysis in a foreseeable future and will provide completely new approaches for interpretation of the multidimensional space? What should we do with the newly found subpopulations that were characterized on the basis of automated processes, but have been neglected so far by manual gating? Do they really exist? What is their biological function? Many questions, and Cytometry Part A will not be able to give answers to all, but to many. It is finally unavoidable to develop corresponding algorithms and more importantly to establish their practical use in the routine laboratory in order to achieve a reliable high-throughput analysis of the complex cytometry data. Here, we regularly have upgrades on the latest software tools for automated data analysis. Last year's Special Section on Automated Image Analysis 1 was edited by Vereb and this year we presented a Special Section on Computational Analysis of Cell Images edited by Rohde which focused on new computer-aided tools for partly automated analysis of microscopic image material for high-throughput data analysis under standardized conditions 2. Because such datasets may contain intrinsic noise, Tyson and coworkers (this issue, page 393) developed a very robust algorithm for automated detection of noisy data sets occurring for example in single molecule DNA–protein experiments. With respect to flow cytometry, several web-based approaches are available for semi-automated analysis 3. Recently, it was reported that FlowGM, a computational pipeline for automated identification of over 20 different cell types, performs with the same precision as manual gating and even improves discrimination of “hard-to-gate” monocyte subsets 4. Totally automated data analysis of CD64 expressing neutrophils as cellular indicator of infection/sepsis correlated perfectly with the manual analysis by a trained cytometrist 5. In contrast to that, semi-automated data analysis can be achieved by the use of the flowDensity R source code that still needs predefined cell populations based on sequential two-dimensional, manual gating 6. So far these innovations appeared somewhat isolated in various issues of the journal, but you can look forward to our Special Issue coming out soon that will be fully dedicated to the plethora of ways for “Computational Analysis of Flow Cytometry Data.” Challenges for automation are of course, that instrument complexity is steadily increasing, bringing new parameters or colors to the user. To fill the gap in the excitation spectrum Zhou and coworkers (this issue, page 419) introduce near-infrared (NIR) flow cytometry. They modified an existing instrument by including an additional NIR channel, enabling 752 nm excitation. In combination with in vivo imaging techniques, where NIR dyes are used, NIR flow cytometry provides the possibility to track NIR fluorescent cells isolated after in vivo imaging from the periphery or the diseased organ in the organism and thereby complementing in vivo imaging. We are very positive that this technique will support development of non-invasive cellular tools, for example for atherosclerosis or cancer detection and could become of high value in individualized medicine. Further development in cytometric technology toward novel diagnostic tools is provided by Rebelo and coworkers (this issue, page 437). The authors compare four different, commercially available instruments that can be modified to detect light depolarization. Many instruments do not offer the possibility to measure depolarized scatter (anymore). But this parameter shows to be relevant for example, in detection of malaria infection as reported by the authors. In infected red blood cells the parasites metabolize hemoglobin into the hemazoin pigment that in turn causes scattered light depolarization 7. So remembering and bringing innovations of the past back is clearly innovative, too.