Multiplex primer prediction software for divergent targets

Gardner, Shea N.; Hiddessen, Amy L.; Williams, Peter L.; Hara, Christine; Wagner, Mark C.; Colston, Bill W.

GeneUp simulates PCR with pairwise combinations of candidate primers which pass length, T m , GC%, and palindrome filters against all target sequences, and uses a greedy algorithm to build a primer set to amplify all of the targets. Since testing all possible pairwise oligonucleotide combinations against each target explodes in time and memory for large target sets, a cap on the maximum number of candidate primers to be tested must be imposed, presumably using the most common oligos, although this is not explicit in the paper. Then a Grey shaded entries indicate where calculations were not run to completion. In other cases (not shaded) where fewer than 100% of targets were predicted to be amplified, the algorithm failed to find primer pairs that met all the required specifications for the remaining targets, that is, primers in the right length, T m , and amplicon length range, with hairpin and dimer avoidance with other primers already selected to be in the set, could not be found. None of the calculations that completed required more than the 16 GB of RAM that was available. PCR simulation (performing a text search for the primer in the target sequence) of each pair versus each target is performed. Unfortunately, the most common oligos may not occur in the correct orientation or distance to serve as primer pairs, so that multiple iterations of the entire process must be performed before a set covering all targets is obtained. MPP, in contrast, uses an efficient ranking algorithm to favor pairs of primer candidates that will produce amplicons of the right size in the most targets. Because of the MPP hashing algorithm and data structures, those targets that are amplified is easily determined without simulating PCR. A copy of the GeneUp software for testing could not be obtained from the authors. The PDA-MS/UniQ approach (26) uses a hash index of 4-mers and scoring heuristic to identify common regions in the target sequences with the most shared tetramers. Simulating combinations of candidate primers selected from these common regions is then performed using a genetic algorithm. Their genetic algorithm is a more scalable approach than those above, and the quality of the solution may be improved with more compute power, although a potential disadvantage of genetic algorithms is that they can be slow, particularly if there are very many possible combinations. For computational tractability, they limited primer size to 12 nt, and avoidance of hairpins and primer dimers was not modeled. This software was not available for download or on a public web server. The other software all required MSA as input. While these tools could not be tested on larger (e.g. family level) primer prediction problems where alignments would not be possible or appropriate, we nevertheless attempted to run these tools on the alignable species-level target sets. The Primaclade (24) webserver timed out for the smallest alignments we tested, Norwalk and FMDV. CODEHOP (23) requires protein alignment as input so is not appropriate for whole-genome (nucleotide) alignments.