An efficient three-Level parallel ABC algorithm for secondary structure prediction of complex RNA sequences

This paper introduces 3L-PABCfold, a parallel artificial bee colony algorithm with three-Level strategy for efficient secondary structure prediction of complex RNA sequences, by multi-processor parallel computing. Designed 3L-PABCfold is a combination of additionally proposed three sub-algorithms: three-Level parallel ABC algorithm (3L-PABC), set based ABC algorithm and discrete ABC (DABC) algorithm. Proposed 3L-PABCfold is an extension of author's earlier developed TL-PSOfold Lalwani et al. (2016), a two-Level particle swarm optimization algorithm for RNA secondary structure prediction. 3L-PABCfold is furthermore enriched with complexity evaluation of highly obscure sequence families followed by makespan minimization and algorithm implementation in parallel computing environment. As the complex dataset, 1,00,000 RNA sequences are randomly generated with length range 139–6358 nucleotides. Level-I of the algorithm reduces the machines idle time by implementing an efficient schedule, whereas, Level-II maximizes the bonded pairs of RNA sequences and Level-III provides optimum secondary structure of RNA sequence. Moreover, three more parallel versions are developed i.e. parallel version of TL-PSOfold (TL-PPSOfold), three Level parallel version of TL-PSOfold (3L-PPSOfold) and two-Level parallel version of 3L-PABCfold (TL-PABCfold), so as to avail a fair comparison and to identify the competent strategy for the addressed problem. The performance evaluated at the criteria of sensitivity, specificity, F-measure along with the descriptive statistics and statistical significance, yields significantly better prediction accuracy for 3L-PABCfold. Further, the efficiency at the processing time criteria proves 3L-PABCfold as the most competent algorithm than TL-PPSOfold, TL-PABCfold and 3L-PPSOfold.