Evaluation of High-Content Screening Fluorescence Correlation Spectroscopic (HCS-FCS) Data

High-content screening fluorescence correlation spectroscopy (HCS-FCS) determines diffusion properties, local concentrations and molecular interactions of biologically relevant proteins in thousands of individual S. cerevisiae cells. We automated one color auto-correlation fluorescence spectroscopy to investigate the properties of the yeast proteome. We also used two-color cross-correlation for interaction screens.For the proteome wide one-color yeast screen, we investigated more than 4000 of the approximately 6000 proteins known in S. cerevisiae. This resulted in over 200,500 measurements on more than 50,000 individual cells. This amount of data required automated tools for data quality control and analysis. Our pipeline started with the evaluation of transmitted light images of the measured cells. We extracted features from these images and trained a support vector machine (SVM) to classify them into healthy yeast cells and samples we wanted to exclude from further analysis. A similar approach enabled us to classify raw fluctuation data taken from the remaining cells and the auto-correlation curves we derived from them. As training sets, we used measured and simulated data.To analyze the correlation curves obtained from the fluctuation data, we fitted different diffusion models. We used mixed-effects models to extract averaged fit parameters for the same protein measured in multiple cells. To select the most appropriate diffusion model we used Akaike's ‘An Information Criterion’ (AIC).This approach not only allows the analysis of large data-sets as they occur in our HCS-FCS experiments and camera based FCS described in literature, but has the additional advantage of reducing the human bias.