Our paper emphasizes the relevance of Extreme Learning Machine (ELM) in Bioinformatics applications by addressing the problem of predicting the disulfide connectivity from protein sequences. We test different activation functions of the hidden neurons and we show that for the task at hand the Radial Basis Functions are the best performing. We also show that the ELM approach performs better than the Back Propagation learning algorithm both in terms of generalization accuracy and running time. Moreover, we find that for the problem of the prediction of the disulfide connectivity it is possible to increase the predicting performance by initializing the Radial Basis Function kernels with a k-mean clustering algorithm. Finally, the ELM procedure is not only very fast but the final predicting networks can achieve an accuracy of 0.51 and 0.45, per-bonds and per-pattern, respectively. Our ELM results are in line with the state of the art predictors addressing the same problem.
Disulfide connectivity prediction with extreme learning machines / Alhamdoosh M.; Savojardo C.; Fariselli P.; Casadio R.. - (2011), pp. 5-14. (Intervento presentato al convegno BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms tenutosi a Roma nel 26/1/2011).
Disulfide connectivity prediction with extreme learning machines
SAVOJARDO, CASTRENSE;FARISELLI, PIERO;CASADIO, RITA
2011
Abstract
Our paper emphasizes the relevance of Extreme Learning Machine (ELM) in Bioinformatics applications by addressing the problem of predicting the disulfide connectivity from protein sequences. We test different activation functions of the hidden neurons and we show that for the task at hand the Radial Basis Functions are the best performing. We also show that the ELM approach performs better than the Back Propagation learning algorithm both in terms of generalization accuracy and running time. Moreover, we find that for the problem of the prediction of the disulfide connectivity it is possible to increase the predicting performance by initializing the Radial Basis Function kernels with a k-mean clustering algorithm. Finally, the ELM procedure is not only very fast but the final predicting networks can achieve an accuracy of 0.51 and 0.45, per-bonds and per-pattern, respectively. Our ELM results are in line with the state of the art predictors addressing the same problem.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.