Counting, measuring, and mapping in FISH-labelled cells: sample size considerations and implications for automation.

Statistical models are used to investigate the need for automation in several potential areas of application of FISH-labelling techniques, including perinatal and tumour cytogenetics, genetic toxicology, and gene mapping. Predictions of the models, based on current estimates of likely error rates for spot-counting and measuring, suggest that a fully automated system is a realistic prospect for detecting full or high-level mosaic trisomies and that interactive systems have the potential to reduce substantially the human workload required to detect residual malignant disease or radiation-induced chromosome aberrations. There appear to be no foreseeable limits to the requirements for speed and accuracy in such systems, since there is effectively no lower limit to the level of relevant biological detail that can be investigated with these techniques.