Predicting failure rate of PCR in large genomes

We have developed statistical models for estimating the failure rate of polymerase chain reaction (PCR) primers using 236 primer sequence-related factors. The model involved 1314 primer pairs and is based on more than 80 000 PCR experiments. We found that the most important factor in determining PCR failure is the number of predicted primer-binding sites in the genomic DNA. We also compared different ways of defining primer-binding sites (fixed length word versus thermodynamic model; exact match versus matches including 1-2 mismatches). We found that the most efficient prediction of PCR failure rates can be achieved using a combination of four factors (number of primer-binding sites counted in different ways plus GC% of the primer) combined into single statistical model GM1. According to our estimations from experimental data, the GM1 model can reduce the average failure rate of PCR primers nearly 3-fold (from 17% to 6%). The GM1 model can easily be implemented in software to premask genome sequences for potentially failing PCR primers, thus improving large-scale PCR-primer design.