AUTOMATED PCR PATHOGEN DETECTION AND QUANTIFICATION

DESCRIPTION (provided by applicant): We will develop software for automated pathogen detection and quantification using data from PCR experiments. Automated pathogen detection using data from a PCR experiment requires software to determine whether DNA from the pathogen is present or absent in a sample. We will develop a pattern-matching algorithm to mathematically analyze PCR amplification data. We will optimize the algorithm against a data set of at least 5000 PCR reactions (including a significant set of data gathered during the anthrax attack) to determine its efficacy and limitations. We expect the pathogen detection algorithms to distinguish positives samples from negative samples in more than 98% of the samples, to find inconclusive results in less than 1% of the samples, and to incorrectly classify less than 1% of the samples. We will also develop software to perform automated melting curve analysis of samples that our detection algorithm has determined to be positive or inconclusive. The melting profile of the probes is a property of the assay, and it can be used for secondary confirmation of a pathogen by comparing the profile of the unknown samples to the profile of the assay's positive controls. We will develop algorithms to automatically determine whether the melting profile of the sample and controls match. With melting analysis confirmation, the failure rate of the final detection algorithm should be less than 0.5%. Automated pathogen quantification requires software to determine the number of copies of a pathogen's DNA in a sample. We will develop discrete dynamical models of PCR for quantification. We will optimize these methods against a large data set of PCR reactions with dilution series. We will systematically determine the features of the models that provide information and the features that can be ignored. We will measure efficacy by comparing computed DNA copy numbers against the known concentrations (as specified by experimenters), and against each other. We will use the most effective model (or models) in the software we produce.