Abstract-Location proteomics is concerned with the systematic analysis of the subcellular location of proteins. In order to perform high-resolution, high-throughput analysis of all protein location patterns, automated methods are needed. Here we describe the use of such methods on a large collection of images obtained by automated microscopy to perform high-throughput analysis of endogenous proteins randomly-tagged with a fluorescent protein in NIH 3T3 cells. Cluster analysis was performed to identify the statistically significant location patterns in these images. This allowed us to assign a location pattern to each tagged protein without specifying what patterns are possible. To choose the best feature set for this clustering, we have used a novel method that determines which features do not artificially discriminate between control wells on different plates and uses Stepwise Discriminant Analysis (SDA) to determine which features do discriminate as much as possible among the randomly-tagged wells. Combining this feature set with consensus clustering methods resulted in 35 clusters among the first 188 clones we obtained. This approach represents a powerful automated solution to the problem of identifying subcellular locations on a proteome-wide basis for many different cell types.

Proteomics seeks the systematic and comprehensive understanding of all aspects of proteins, and location proteomics is the relatively new subfield of proteomics concerned with the location of proteins within cells. This review provides a guide to the widening selection of methods for studying location proteomics and integrating the results into systems biology. Automated and objective methods for determining protein subcellular location have been described based on extracting numerical features from fluorescence microscope images and applying machine learning approaches to them. Systems to recognize all major protein subcellular location patterns in both two-dimensional and three-dimensional HeLa cell images with high accuracy (over 95% and 98%, respectively) have been built. The feasibility of objectively grouping proteins into subcellular location families, and in the process of discovering new subcellular patterns, has been demonstrated using cluster analysis of images from a library of randomly tagged protein clones. Generative models can be built to effectively capture and communicate the patterns in these families. While automated methods for high-resolution determination of subcellular location are now available, the task of applying these methods to all expressed proteins in many different cell types under many conditions represents a very significant challenge.

murphy@cmu.edu.

We have previously built a Subcelluar Location Image Finder (SLIF) system, which extracts information regarding protein subcellular location patterns from both text and images in journal articles. One important task in SLIF is to identify fluorescence microscope images. To improve the performance of this binary classification problem, a set of 7 edge features extracted from images and a set of "bag of words" text features extracted from text have been introduced in addition to the 64 intensity histogram features we have used previously. An overall accuracy of 88.6% has been achieved with an SVM classifier. A co-training algorithm has also been applied to the problem to utilize the unlabeled dataset and it substantially increases the accuracy when the training set is very small but can contribute very little when the training set is large.