Proportional fault-tolerant data mining with applications to bioinformatics

Area of Science:

• Bioinformatics
• Data Mining
• Computational Biology

Background:

• Frequent pattern mining is crucial for extracting insights from large datasets.
• Existing fault-tolerant (FT) data mining methods are limited for noisy real-world data, especially in biological databases.
• Biological data often contains noise, necessitating robust pattern mining techniques like FT approaches.

Purpose of the Study:

• To propose and develop a novel concept of proportional fault-tolerant (FT) pattern mining.
• To design algorithms capable of mining patterns where tolerable faults are proportional to pattern length.
• To demonstrate the efficacy of proportional FT mining on real biological data.

Main Methods:

• Introduced the concept of proportional FT pattern mining, where fault tolerance scales with pattern length.
• Developed two algorithms: FT-BottomUp (using FT-Apriori heuristic) and FT-LevelWise (grouping patterns by fault tolerance).
• Applied the proposed algorithms to real biological datasets, including SARS-CoV spike protein data.

Main Results:

• Successfully identified two reported epitopes of SARS-CoV spike proteins within the mined itemsets.
• Demonstrated that proportional FT data mining is more effective than fixed FT data mining for the studied application.
• The FT-BottomUp and FT-LevelWise algorithms effectively mine proportional FT patterns.

Conclusions:

• Proportional FT pattern mining offers a more suitable approach for analyzing noisy biological data compared to fixed FT methods.
• The developed algorithms provide efficient solutions for discovering biologically relevant patterns with adjustable fault tolerance.
• This technique has significant potential for applications in bioinformatics, such as motif discovery and epitope identification.