SPADA Draft Documents

497

498

Multiplexing is a complex system

Multiplex PCR is a complex system with many interacting variables (e.g., multiple primers, 499 genomic DNA from the target(s) of interest, and contaminating genomes from the sample 500 matrix). Optimizing one set of primers can result in an unexpected interaction with another set of 501 primers in the reaction to create unexpected false amplicons from the exclusivity and background 502 databases or primer-amplicon cross-hybridization reactions. In addition, as the size of the 503 multiplexed reaction grows larger, there is a combinatorial explosion in the permutations of 504 possible multiplex reactions. Consider a 20-plex reaction with 10 primer design candidates (i.e. 505 the top scoring singleplex designs) for each of the 20 targets: the number of possible multiplex 506 reactions to consider is 10 20 . An iterative empirical approach can only assess a few reactions 507 (without knowledge of the interacting variables) and thus samples a very limited amount of the 508 sequence design space, almost always resulting in a sub-optimal design and often complete 509 failure. Even an iterative brute-force computational approach using high-performance computing 510 that could test millions of combinations of multiplex sets will vastly under sample the number of 511 possible reactions. For the 20-plex example with 10 candidate solutions, it would be 512 computationally intractable to run thermodynamic scanning (e.g., PrimerBLAST or 513 ThermoBLAST ) for all 10 20 possible multiplex permutations to ensure that false positives are 514 minimized. Thus, a modern sophisticated algorithm is needed to solve the combinatorial 515 explosion thoroughly and find the optimum multiplex design. The program PanelPlex uses a 516 depth-first search with a pruning algorithm to accomplish the multiplexing design task and 517 guarantees finding the optimum solution (or top N solutions) out of the entire space of possible 518 multiplexes. We recommend using such multiplex optimization algorithms to greatly reduce the 519

28