The Department of Computer Science at the University of Cyprus cordially invites you to the Seminar entitled:

Developing of Novel and Efficient Second-Order Learning and Optimisation Techniques for Protein Structure Prediction

Speaker: Michalis Agathocleous and Petros Kountouris
Affiliation: University of Cyprus, Cyprus
Category: Seminar
Location: Room 148, Faculty of Pure and Applied Sciences (FST-01), 1 University Avenue, 2109 Nicosia, Cyprus (directions)
Date: Thursday, 24 November 2011
Time: 16:30-17:30 EET
Host: Chris Christodoulou (cchrist AT and Vasilis Promponas (vprobon AT

Protein secondary structure prediction (PSSP) is a critical step towards modelling protein 3D structure and, potentially, protein function. Over the past 20 years, machine learning techniques and evolutionary information have significantly boosted the quality of PSSP methods. However, there is still an opportunity for more accurate prediction through the use of more sophisticated learning algorithms. The Bidirectional Recurrent Neural Network (BRNN) architecture of Baldi et al. [Baldi et al., Bioinform., 15, 937-946, 1999] is currently considered as one of the most optimal computational neural network type architectures for addressing the problem. In this project, we have implemented the same BRNN architecture, but we have used a modified training procedure [Agathocleous et al., IFIP International Federation for Information Processing AICT, 339, 128-137, 2010]. More specifically, our aim is to identify the effect of the contribution of local versus global information, by varying the length of the segment on which the Recurrent Neural Networks operate for each residue position considered. For training the network, the backpropagation learning algorithm with an online training procedure is used, where the weight updates occur for every amino acid, as opposed to Baldi et al., where the weight updates are applied after the presentation of the entire protein sequence. The performance of our method improved even further through the use of an ensemble of BRNNs. In addition, we have developed and implemented the Scaled Conjugate Gradient optimization algorithm to train the BRNN architecture. Additionally, our work focuses on the challenging problem of filtering PSSP, a common final step in many PSSP methods which aims to smooth the results and provide physicochemically realistic predictions [Kountouris et al., IEEE/ACM Trans. on Comput. Biol. and Bioinform., submitted, 2011]. Despite being employed widely, to the best of our knowledge, no study has been carried out to find the most suitable filtering technique. Herein, we perform a comparative study on the challenging problem of filtering, utilising both empirical smoothing rules and machine learning techniques.

Short Bio:
Michalis Agathocleous received a B.Sc. degree (with distinction) in Computer Science from the University of Cyprus, Nicosia, Cyprus, in 2009, and a M.Sc. degree in Machine Learning, from the University College London (UCL), London, U.K., in 2010. He is currently pursuing a Ph.D. degree at the Department of Computer Science, University of Cyprus. His current research interests include bioinformatics, neuroscience, machine learning and computational intelligence. Petros Kountouris initially obtained a B.Sc. in Computer Engineering and Informatics from the University of Patras, Greece, in 2006. He later moved to the UK where he received a Ph.D. in Chemistry/Bioinformatics from the University of Nottingham in 2010. He is currently working as a post-doctoral researcher at the Department of Computer Science in the University of Cyprus, Cyprus. He has continuous interest for structural computational biology and on possible applications of computational intelligence and machine learning to bioinformatics.

  Other Presentations Web:
  Colloquia Web: