Quantification of subcellar localization and co-localization from high-content images

Abstract

The knowledge of subcellular localization and co-localization of proteins is crucial for investigating how proteins function and interact with each other within a cell. Plenty of research groups have dedicated their efforts in characterizing and predicting the subcellar localizations of proteins. In this thesis, i introduce the fluorescence microscopic term - co-localization into the process of quantification and prediction of subcellar localization, and the corresponding computational frameworks are developed to perform and evaluate the performance of the co-localization in quantifying and predicting of subcellular localization. The co-localization measurement can either work as a single parameter to estimate the relationship between a specific protein and a series of subcellular compartments to generate a co-localization profile; or work with many other statistic measurements serving as a set of co-localization features, which can be used to predict the subcellular localization itself or with other protein sequence features to improve the prediction accuracy. On the other hand, tested with synthetic images, we developed a method based on the co-localization measurement to estimate the co- occurrence and correlation separately, which would be greatly helpful in providing biological meaningful explanations to the quantification results, especially for mid-value results. The methods were validated and applied on 2-D cytoskeletal protein image ataset and 3-D transcription factor image dataset.