|
|
|
Round-the-clock imaging provides insight to
intracellular macromolecular changes
and sub- cellular localization.
Cytomics refers to the study of cell phenotype and function.
 |
|
Robotics
A robotic microscopic stage and bioenvironment allow for cell culture system with dynamic imaging.
Time-lapsed images are obtained at user-defined intervals (as frequent as 1 image per second).
Visible and epiflourescnce imaging are possible using the Nikon Eclipse TE 2000 U microscope and
Photometric ES Cool Snap CCD camera. The platform software further allows for combinatorial experimental
design where more than 350 locations or treatments can be studied during one experiment.
The bioenvironment is controlled at 37 C, 5% CO2 and humidified.
Image Processing and Database
Image analysis algorithms allow for quantitation of a large number of parameters for individual cell characteristics and behaviors.
There are more than 65 LCI quantitative measurements, among these: 1) proliferation: live cell number, total cell number,
apoptotic cell number1; 2) individual cell short-term and long-term motility: curvilinear, straight-line, angular, time-smoothed,
and instantaneous velocities as well as persistence and speed2-5; 3) individual cell morphology: area/size, length, breadth,
best-fit ellipse, number of pseudopodia6; 4) individual cell fluorescent molecular expression in up to 6 wavelengths,
assigned to various probes; and 5) in vivo performance outcome measures (see Appendix). Imaging data can also be combined
with flow cytometry data and other biologic measures to provide further breath to the cell population profile.
Unique comprehensive phenotype profiles can be correlated with other behaviors or with in vivo behaviors.
We subsequently perform cytomic analysis and comparative informatics on cell populations of interest.
Cytomics refers to the study of cell phenotype and function. Populations may also be analyzed by flow cytometry which
to provide an in vitro molecular phenotype.
Informatics
We use a multidimensional relational database (RDBMS) to manage the large datasets which includes tables for all
measurements obtained from time-lapsed data obtained from image processing, and also tables for associated measurements by
FACS or in vivo measures. Database queries then associate the measured parameters with corresponding biological and experimental
factors, facilitating the analysis. The relational database was built using Microsoft SQL server 2005.
A number of statistical and machine learning methods are utilized. Exploratory data analysis identifies confounding nuisance
and blocking factors. Feature selection by multivariate analysis and machine learning (Bayes-based and Support-Vector machines)
allows us to focus on the parameters which best describe the differences between the study populations.
We use machine learning (mixed-models clustering) to detect cell subpopulations. We compare the dynamics among the different
cell populations using mixed-effects models, a generalization of multi-way ANOVA. A number of other robust statistical approaches
are employed including robust model fitting, permutation tests, post-hoc effect size analysis, and cumulative sum control charts.
Software engineering includes the RDBMS, as described above, and use of MatLab and R-project for statistics and machine learning.
We create software programs that generate experimental reports automatically; this aids in the repeatability of the analysis with
newer data as they arrive. We use the visual environments of Spotfire, SPSS and R-project for the exploratory data analysis for the
cell behavioral phenotype. Trellis plots and parallel coordinate plots are used to identify similarities (or differences) in the
cell populations as the populations grow (and change) over time. Graphical displays similar to microarray display, or like the one
shown here, are created for the comprehensive cytomic data.
|
