Elsevier

Expert Systems with Applications

Volume 36, Issue 10, December 2009, Pages 12210-12213
Expert Systems with Applications

Location of amide I mode of vibration in computed data utilizing constructed neural networks

https://doi.org/10.1016/j.eswa.2009.04.065Get rights and content

Abstract

An automatic location method of amide I mode of vibration between computed data is proposed, based on the well established neural network model of artificial intelligence. This method was developed by constructing and testing the neural network on previously computed and characterized data which were divided in the training and the testing set, respectively. The results show high level of success since the majority of amide I modes of vibration in the testing set were located 99.5%.

Introduction

Infrared spectroscopy is a powerful tool for the identification of characteristic groups which are present in the molecule under study (McMurry, 1996). The IR region of the spectrum comprises decomposite information of the investigated molecular system which extended the use of IR spectroscopy; in addition to the primary structure determination of a small molecule, IR spectroscopy can be a valuable tool for the determination either of the secondary structure of a protein or for many current research fields related to biological phenomena (Barth, 2007).

A special chapter of the infrared spectroscopy of proteins and polypetides is the extensive study of the amide vibrational modes either by experimental procedures (Barth, 2007, Ganim et al., 2008) or by theoretical simulations (Schultheis, Reichold, Schropp, & Tavan, 2008). Both studies have revealed significant details concerning the understanding of the structure and function of these molecules of life. The amide I mode along with the amide II and amide III modes provide distinct bands of high absorbance in the IR spectrum.1

Most computational packages have embodied the quantum calculation of normal modes of a molecule. Among them, the most cited packages Gaussian (Frisch et al., 1998) and Gamess (Schmidt et al., 1993) in the field of computational chemistry. A number of visualization packages (Humphrey et al., 1996, Schaftenaar and Noordik, 2000) can read the output of the computation of these packages and visualize each normal mode, adjusting a number of parameters such as the amplitude and the number of frames per animation cycle during the visualization. These packages, finally provide the user with a clear animated picture of the normal mode comparable to a number of factors. Despite the enormous work on the visualization packages, up to date, to our knowledge, there is no package available for the automated location and recognition of amide I, II and III modes of a peptide or protein.

Neural networks (Bishop, 1995, Cybenko, 1989) are well-established artificial intelligence tools known for their approximation capabilities. They have been used successfully in many scientific fields such as pattern recognition (Artyomov & Yadid-Pecht, 2005), signal processing (Uncini, 2003), astronomy (Valdas & Bonham-Carter, 2006), solution of ordinary and partial differential equations (Lagaris, Likas, & Fotiadis, 1998). They have been also used for medical diagnosis and prediction (Burke et al., 2000, Eggers et al., 2007). In addition, neural networks have been tested successfully on modern biological problems such as backtranslation of amino acid sequences (White & Seffens, 1998) and prediction of the protein secondary and tertiary structure (Böhm, 1996).

In this work we applied constructed artificial neural networks to a computed data set of normal modes belonging to a selected group of molecules of interest, such as amides, dipeptides and an oligopeptide. The aim was to establish an automatic location method initially for of the amide I mode as this is outputted by a computational chemistry package. The package used in this case was the Gaussian98 but the method can be applied to any computational package capable of producing the simple but necessary Cartesian format of the output.

The rest of this article is organized as follows: in Section 2 we give a brief description of the proposed method, in Section 3 we list the experimental results and in Section 4 we discuss them and we propose specific enhancements of the present method.

Section snippets

Method outline

In this section we give a brief description of the neural networks and we present the steps of the proposed method.

Experimental results

We applied the neural network construction method 30 times on the train set, using different seed for the random generator each time. Every constructed neural network were applied to the test set. The number of chromosomes used in the construction method was set to 500 and the maximum number of generations allowed was set to 200. The crossover rate was set to 0.1 and the mutation rate to 0.05. The average training error was 0.23% and the average testing error was only 0.55%, while the average

Discussion

From the conducted experiments we can conclude that:

  • The method of the construction of neural networks can be applied with great success to the general problem of the location and characterization of discrete oscillations of a group of bodies (atoms) with a variety of different sets of velocities. Amide I mode and its location among the computed set of the molecular normal modes resembles this motion.

  • The method is capable of constructing and training small and compact neural networks with good

References (26)

  • Frisch, M. J., Trucks, G. W., Schlegel, H. B., Scuseria, G. E., Robb, M. A., Cheeseman, J. R., et al. (1998)....
  • Z. Ganim et al.

    Amide I two-dimensional infrared spectroscopy of proteins

    Accounts of Chemical Research

    (2008)
  • I.E. Lagaris et al.

    Artificial neural networks for solving ordinary and partial differential equations

    IEEE Transactions on Neural Networks

    (1998)
  • Cited by (0)

    View full text