On characterizing protein spatial clusters with correlation approaches

Spatial aggregation of proteins might have functional importance, e.g., in signaling, and nano-imaging can be used to study them. Such studies require accurate characterization of clusters based on noisy data. A set of spatial correlation approaches free of underlying cluster processes and input parameters have been widely used for this purpose. They include the radius of maximal aggregation r(a) obtained from Ripley's L(r) - r function as an estimator of cluster size, and the estimation of various cluster parameters based on an exponential model of the Pair Correlation Function(PCF). While convenient, the accuracy of these methods is not clear: e.g., does it depend on how the molecules are distributed within the clusters, or on cluster parameters? We analyze these methods for a variety of cluster models. We find that ra relates to true cluster size by a factor that is nonlinearly dependent on parameters and that can be arbitrarily large. For the PCF method, for the models analyzed, we obtain linear relationships between the estimators and true parameters, and the estimators were found to be within +/- 100% of true parameters, depending on the model. Our results, based on an extendable general framework, point to the need for caution in applying these methods.