Proportional fault-tolerant data mining with applications to bioinformatics

The mining of frequent patterns in databases has been studied for several years, but few reports have discussed for fault-tolerant (FT) pattern mining. FT data mining is more suitable for extracting interesting information from real-world data that may be polluted by noise. In particular, the increasing amount of today's biological databases requires such a data mining technique to mine important data, e.g., motifs. In this paper, we propose the concept of proportional FT mining of frequent patterns. The number of tolerable faults in a proportional FT pattern is proportional to the length of the pattern. Two algorithms are designed for solving this problem. The first algorithm, named FT-BottomUp, applies an FT-Apriori heuristic and finds all FT patterns with any number of faults. The second algorithm, FT-LevelWise, divides all FT patterns into several groups according to the number of tolerable faults, and mines the content patterns of each group in turn. By applying our algorithm on real data, two reported epitopes of spike proteins of SARS-CoV can be found in our resulting itemset and the proportional FT data mining is better than the fixed FT data mining for this application.