Data and Image Analysis Special Interest Group Meeting,
SBS 10th Conference – Orlando, Florida, September 13, 2004
In spite of some confusion regarding the date and time of the meeting, the word apparently got out, and approximately 50 people participated in the meeting.
The first item of business was a very brief review of the very brief activity over the last year. Essentially no activity had taken place, and there had been very little traffic on the Yahoo group devoted to discussion of issues relevant to this SIG. The main action point from last year’s meeting was to set up a vehicle to share best practices in dose-response data analysis. This still needs to be done. It was generally agreed that more effort needed to be put into collecting various bits of reference material on data analysis and placing it on the SIGs website. The topic of creating an XML-based standard for plate readers was again brought up for the benefit of new participants; again, no interest was generated.
Next, Ilya Ravkin of Vitra Bioscience made a quick presentation of a simple analysis of the content of posters and presentations at this year’s SBS meeting. According to this analysis, there were more than twice as many posters and presentations dealing with some aspects of image analysis or with high content assays versus “ordinary” numerical data analysis. Ilya made the point that this fact must be recognized in the structure of SBS SIGs. One way would be to create a new SIG devoted specifically to this topic, at the risk of diluting and drawing people away from the existing DA/IT SIG. Alternatively, the remit of the DA/IT SIG should be expanded to include this topic. After some discussion, it was decided that the better course, at least for the time being, was to expand the remit of the DA/IT SIG to include data analysis issues specific to high content screening. If this topic takes over the discussion in the DA/IT SIG, or if members not doing HCS or image analysis feel left out, the issue of creating a new SIG will be raised again at that time. This issue will be on the agenda at next year’s meeting.
The technical topic for this year’s meeting was a panel discussion on how data analysis of high content assays differs from data analysis in traditional screening, and whether new tools and metrics were needed. Four panelists had been selected, each of which had made either a poster or oral presentation at this SBS meeting on some topic relevant to high content data analysis. Each panelist began by giving a short summary of his or her poster or presentation. The panelists and their topics were as follows:
Ilya Ravkin of Vitra Bioscience. Ilya presented data showing that the Z and Z’ metrics currently used so widely in screening data analysis could be affected by image analysis to make an assay appear to have a higher diagnostic value than it actually had and suggested taking the whole dose curve into account for determining assay quality.
Liz Roquemore of GE Healthcare. Liz reviewed her award-winning poster of last year in which she also showed that Z and Z’ were not necessarily the best assay metrics, and described some other metrics which might be more robust.
Mark Collins of Cellomics. Mark reviewed some of the techniques used in the Cellomics software to extract more information from images to better reject artifacts.
Søren Møller of BioImage. Søren presented the results of a re-analysis of a large HCS screening campaign at BioImage using data mining techniques, demonstrating that more information could be obtained from the existing image analysis result data, and that it was possible to reject more artifacts and make better decisions regarding which compounds to carry forward.
The discussion centered on two main topics:
Statistical validity of Z measures; their proper and intended use and possible subjective factors.
General considerations about quality measures for image-based assays. This area is the interface between image analysis with its traditional quality assessment as correspondence with manual analysis and high throughput screening with its quality measures based on statistics of the defined assay states. The clash of these two approaches is generating many interesting and fruitful ideas.
Over the coming year we plan to discuss data interchange standards and comparison of image analysis algorithms for some important assays as well as any other topics that may come up in the interim. These topics will form the agenda for the next year’s DA/IT SIG meeting.
Finally, on behalf of the DA/IT SIG, I would like to thank Liz, Mark, Ilya, and Søren for their participation in the panel discussion. I would also like to thank Ilya for his help in organizing this year’s meeting.
Kurt Scudder, Matrical, Inc., SIG chairman