Predicting protein stability change upon double mutation from partial sequence information using data mining approach

The prediction of stability change for protein mutants is one of the important issues in protein design. Recently, the prediction upon double mutation has attracted more and more attention. In this work, we have employed a data mining approach to discriminating stability change for protein double mutants. We incorporated a reliable rule induction algorithm along with accuracy of 82.2% to construct rule-based knowledge patterns. Further, a fuzzy query method was utilized to value important and similar rule patterns for an input with partial sequence information. The results showed that the approach has two major advantages: (i) A rule-based knowledge representation offers intuitive interpretation on raw data, which is helpful to understand the content; and (ii) A fuzzy query method incorporates the concept of uncertainty, which can make predictions from partial information. Based on the proposed approach, we have also developed a web service for predicting protein stability change upon double mutation from partial sequence information and it is available at http://bioinformatics.myweb.hinet.net/tandem.htm.