International Journal of Data Mining and Bioinformatics
| Hi-index | 0.01 |
This paper addresses an important and vital problem within the general area of disease recognition, namely identifying disease biomarker genes. Given the complexity of this domain, the basic idea tacked in this paper is employing multiple agents to handle this problem. Though the developed methodology is general enough to be applied to any other domain, we concentrate on identifying cancer biomarkers in this paper. Our approach is mainly based on detecting the minimum set of genes that could successfully identify cancer samples. Multiple agents are involved in the process. After each agents applies its own rules and reports candidate cancer biomarkers, the agents negotiate to agree on the actual biomarkers. The latter process may require further investigation of the characteristics of each of the reported genes because some of them may have the same functionality and the target is a compromise of the best representative of each functionality. A degree of confidence in each candidate biomarker influences the negotiation process. The so far conducted experiments reported very encouraging results with high classification rate; none of the involved agents could alone achieve a close success~rate.