Modeling the Amplification of Immunoglobulins through Machine Learning on Sequence-Specific Features

Abstract

Successful primer design for polymerase chain reaction (PCR) hinges on the ability to identify primers that efficiently amplify template sequences. Here, we generated a novel Taq PCR data set that reports the amplification status for pairs of primers and templates from a reference set of 47 immunoglobulin heavy chain variable sequences and 20 primers. Using logistic regression, we developed TMM, a model for predicting whether a primer amplifies a template given their nucleotide sequences. The model suggests that the free energy of annealing, ΔG, is the key driver of amplification (p = 7.35e-12) and that 3′ mismatches should be considered in dependence on ΔG and the mismatch closest to the 3′ terminus (p = 1.67e-05). We validated TMM by comparing its estimates with those from the thermodynamic model of DECIPHER (DE) and a model based solely on the free energy of annealing (FE). TMM outperformed the other approaches in terms of the area under the receiver operating characteristic curve (TMM: 0.953, FE: 0.941, DE: 0.896). TMM can improve primer design and is freely available via openPrimeR (http://openPrimeR.mpi-inf.mpg.de).