Protein Secondary Structure Prediction Using FFA Optimized ANN
1Durairaj M, 2Sivakumar S, 3Sangeetha B, 4Saravannan K, 5Saravanakumar K
1Department of Physics, Mahendra Engineering College, Mallasamudram West, Namakkal, Tamil Nadu, INDIA
2Department of Physics, Government Arts College (Autonomous), Salem, Tamilnadu, INDIA
3Department of EEE, AVS Engineering College, Salem, Tamilnadu, INDIA
3Department of Physics, AVS Engineering College, Salem, Tamilnadu, INDIA
5Department of Physics,Mahendra Institute of Technology, Namakkal,Tamilnadu, INDIA
ABSTRACT
The epidermal growth factor (EGF) family of RTKs plays a vital role in regulation of cell proliferation, differentiation, survival and migration. It carries both restricted and redundant function during developed stage of mammalian and maintains tissues at the adult stage. While the regulation carried by these receptors gets decreased, it will lead to many diseases like cancer among humans. Hence, the understanding about the function and regulation of RTK is essential for the development of drugs for human diseases. As the SS of a protein is responsible for the interactions among proteins, it is difficult for the scientists to understand their mutual relationships and functions.
Hence the prediction of PSS is considered as a difficult task. Even though PSO based optimization topology exhibits higher accuracy in PSS prediction there may be some drawbacks, when it is subjected to high-dimensional space. Its convergence rate under high-dimensional space is very low. Hence to overcome this drawback, this work utilized a nature inspired firefly algorithm to tune ANN. It will results in improved convergence rate and also consumes less time with high accuracy.
Keywords
Protein Secondary Structure (PSS);Firefly; ANN; RTK
Introduction
Growth factors are essential for the growth, development and homeostasis of multicellular organisms. These growth factors are essential for cell to cell communications underlying embryonic tissue induction, cell survival, fate determination, tissue specialization, apoptosis and cell migration. These receptors transduce extracellular signals through the activation of intracellular messengers or through receptor translocation to the nucleus. The epidermal growth factor (EGF) family of RTKs is also referred as ErbB or HER receptors [1].
Hence, it is most widely studied for its role in development and physiology.
BPNN can be utilized in many applications. It relies on bias, weight and learning algorithm adopted in a NN design [2-8]. In this, steepest descent algorithm is normally implemented in BPNN topology. It is also called gradient method. In this topology, convergence time is key factor. Conjugate BP and LM BPNN is normally utilized to reduce the convergence time. As conjugate utilizes second order derivative, it converges with a less number of iterations[9,10].
LM BP is the standard topology, which is based on the nonlinear least square algorithm[11-13]. However, performance of these algorithms is high it needs a memory for optimization. Hence, numerous topologies is adopted to tune ANN.
The recent development in optimization topologies is extremely helpful to solve complex problems. They can also be implemented in the form of hybrid nature and are adopted as standard algorithm.
Among various types of algorithms GA plays a vital role. Thus, BPNN associated with GA will provide best solution and hence, it is named as GA BPNN. It performs well in any dynamic nature of environment. However, it exhibits slow convergence because of large search space [14-18].
In all the optimization procedures, weights, bias are updated periodically. Hence, this work proposed a novel nature based topology to optimize the function of ANN. Thus, the proposed FFA will exhibit convergence with short duration of time.
It is also observed that this proposed topology is never confined to a local minimum, and it is completed quickly if the convergence condition of a data set is already predefined.
FFA
In this topology, flies utilize the flash signals to fascinate other flies for mating. The movement of firefly would be random if there are no other insects in its proximity and its movement. This is because of light intensity of the firefly. The intensity (L) of light decreases as the distance (r) increases and thus most flies can communicate only up to several hundred meters. Thus, the formulated fitness function is always associated with the brightness of flash[19-22].
Accordingly, the intensity of the flashing light is calculated as 𝐿𝑖 = 𝐿𝑜𝑒 −𝑛𝑑2 (1) The distance between two flies can be calculated as
𝑑𝑖𝑗 = 𝑓𝑖 − 𝑓𝑗 = 𝑓𝑖 − 𝑓𝑗 2 (2)
Thus by modifying the flash light absorption parameter, a quick convergence can be obtained. But in most of the cases it is fixed. Initially, any one of the fly is considered as a brightest one and rest of the flies are moved towards to that.
During this process, the distance and attractiveness of every firefly with respect to the brighter one is estimated. Finally, the flies are arranged on the basis of their performance.
Thus, the procedure adopted by FFA ANN topology is depicted in Figure 1.
Figure 1.FFA ANN topology
The proposed FF based BPNN converges with less number of iteration. The number of iteration and the rate of correct classification can be enhanced by increasing the population.
Study Area and Data
In this work, 100 proteins set for training and 5 protein set for testing were used. All these sets have 3 secondary structure classes (α-helix/ β-strand/coil). Each set was utilized as the validating and testing set. The parameter configuration used for FFA is depicted in Table 1.
Table 1. Parameter configuration of FFA
Parameters Values
No of parameter 2
Range of the chosen parameter -10 to 10
No. of Fireflies 50
No.of Iterations 200
0.81
1.00
0.98
Results and Discussion
Among the training data set 47% were about coil, 31% were helix and 21% were strand.
In the testing data set, it was about 48% of C, 31% of H and 21% of E.
Figure 2.Best validation performance curve
The MSE obtained for an ANN model is depicted in Figure2. Thus from the above figure it is found at 62 epoch, the best validation is about 0.29. Thus, it is concluded that as the error of the network gets reduced, the efficiency is high.
Thus, the predicted secondary for the sequence of chain A of RTKs using FFA tuned NN is tabulated below.
Table 2.Predicted secondary structure of RTKs under different topology
Methods Secondary Structure
Sequence (1-50) GEAPNQALLR ILKETEFKKI KVLGSGAFGT VYKGLWIPEG EKVKIPVAIK
Structure
DSSP **S******* B**GGG***E EEEEE*SS*E EEEEEE**TT S**BEEEEEE MLNN CCCCCHHHHH HHHHHHHCCC CCCCCCCCCE EEEEEECCCC CCCCEEEEEE PSONN CCCHHHHHHH HHHHHHHCCC EEEEECCCCC EECCEEECCC CEEEEEEEHH Proposed
FFANN CCCCHHHCHH CCCHHHEEEE EEECCCCCCE EEEEEEECCC CCCEEEEEEE Sequence(51-100) ELREATSPKA NKEILDEAYV MASVDNPHVC RLLGICLTST VQLITQLMPF
Structure
DSSP **SSTT*THH HHHHHHHHHH HTT***TTB* *EEEEEESSS EEEEEE**TT MLNN HHHCCCCCCC HHHHHHHHHH HHHCCCCCHE EEEEEECCCC HHHHHHHHHH PSONN HHCCCCCHHH HHHHHHHHHH CCCCCCCCEE EEEEEEECCC EEEEEHHHCC Proposed
FFANN EECCCCCHHH HHHHHHHHHH HHHCCCCCEE EEEEEEECCC CEEEEECCCC Sequence(101-150) GCLLDYVREH KDNIGSQYLL NWCVQIAKGM NYLEDRRLVH RDLAARNVLV
Structure
DSSP *BHHHHHHHH TTT**HHHHH HHHHHHHHHH HHHHHTTEE* S***GGGEEE MLNN HHHHHHHHHC CCCCCHHHHH HHHHHHHHHH HHHHHHHHHH HHHHHHHEEE PSONN CCHHHHHHHC CCCEEEEHHH HHHHHHHHHH HHHHHHCHHH HHHHHHHHHH Proposed
FFANN CCHHHHHHHH CCCCCHHHHH HHHHHHHHHH HHHHHHCCCH HHHHHHHHHE Sequence(151-200) KTPQHVKITD FGLAKLLGAE EKEYHAEGGK VPIKWMALES ILHRIYTHQS
Structure
DSSP EETTEEEE** *TT*EESS** ********** **GGG**HHH HHS****HHH MLNN ECCCCEEEEH HHHHHHHCCC CCHEEHCCCC CCEEEECHHH HHHHCCCCCC PSONN CCCCEEEEEC CCCEEEECCC CCCCCCCCCC EEEECCCHHH HHHHHCCCCC Proposed
FFANN CCCCEEEEEC CCCCEECCCC CCEEECCCCC EEEECCCHHH HHHCCCCCHH Sequence(201-250) DVWSYGVTVW ELMTFGSKPY DGIPASEISS ILEKGERLPQ PPICTIDVYM
Structure
DSSP HHHHHHHHHH HHHTTS**TT SSS*GGGHHH HHHHT***** *TTB*HHHHH MLNN CHEEEEEEHH HEEECCCCCC CCCCCHHHHH HHHCCCCCCC CCCCCHHHHH PSONN CCEEEEEEEE ECCCCCCCCC CCCCCHHHHH HHHHCCCCCC CCCCHHHHHH Proposed
FFANN HHHHHHHHHH HHHHCCCCCC CCCCHHHHHH HHHHCCCCCC CCCCCHHHHH Sequence(251-300) IMVKCWMIDA DSRPKFRELI IEFSKMARDP QRYLVIQGDE RMHLPSPTDS
Structure
DSSP HHHHHT*SSG GGS**HHHHH HHHHHHTTSH HHHB**TT*S S*********
MLNN HHHHHHCCCC CCCCCHHHHH HHHHHHCCCC CCEEEEECCC CCCCCCCCCC PSONN HHHHHHHHCC CCCCHHHHHH HHHHHHHCCC HHHHHCCCCC CCCCCCCCCC Proposed
FFANN HHHHHHHCCH HHCCCHHHHH HHHHHHHHCC HHHCCCCCCC CCCCCCCCCC Sequence(300-327) NFYRALMDEE DMDDVVDADE YLIPQQG
Structure
DSSP ********SS **TTB**TTT *******
MLNN HHHHHHHHHH HHHHHHHHHH HHHCCCC PSONN CCCCCCCCHH HHHHHHHHCC CCCCCCC Proposed
FFANN CCCCCCCCCC CCCCCCCCCC CCCCCCC
From the above analysis, the predicted H is about 40.36%and E is about 17.43%, hence the proposed RTKs is of mixed class rather than all helix or beta.
The impact of AA’s composition in PSS is investigated. This is displayed in terms of 3 type SS. Similarly, the effect of physiochemical property of AA over SS prediction is also investigated here.
It can be seen that AAs are not present in uniform quantities in SS. There are very few AAs such as Glutamine andLeucine and have the highest number of helix residues while Alanine and Serine have the exact number of helix residues in it. In the case of strand, Lysine, and Leucine seem to have the highest content while most of the acids like Asparagine, Aspartic acid does not exist strand. Similarly, large numbers of residues of coil are in Glutamic acid, Leucine and while Alanine, Asparagine seem to have the same content in the coil and the same is depicted in Figure 3.
Figure 3.Content of AAs in SS
Figure 4.Percentage of SS in an AA
From the above Figure 4, it is concluded that all AAs exhibit helix nature and AAs such as Asparagine, Aspartic acid have not exhibit strand nature. Thus, it is concluded that -helix plays a vital role in 2J5F protein.
A lot of measurements are adopted to examine the accuracy of prediction topology. Q3 accuracy represents the correctly predicted state in percentage. Thus, the Q3 accuracy is represented as QH, QE and QC[23].
SOVprovides the overlap of predicted 3 SS by observing the predicted and measured segments [24,25] and is shown in Table 3.
Table 3.Performance analysis of Q3 and SOV of the proposed topology Q overall (%) QH (%) QE (%) QC (%)
Q3 67.9 95.4 32.7 77.1
SOV 53.8 95.6 31.7 42.0
Comparison with Other Methods
Finally, the results obtained using this proposed methodology is compared with the performance of other networks which depicted the SS of RTKs.
Table 4.Comparative analysis of performance of the other methods in secondary structure prediction
Method Alpha (%) Beta sheet(%) Coil (%)
DSSP 33 15 -
STRIDE 35 16 -
MLNN 38.53 16.20 45.25
PSO-NN 38.83 19.26 31.19
FFA-NN 40.36 17.43 42.21
Figure 5.Predicted secondary structure
From the Table 4 and Figure 5, it is concluded that the proposed FFA-NN gives more and better prediction of SS than other topologies.
Table 5.Performance comparison of Q3 and SOV of the proposed topology Qoverall (%) QH (%) QE (%) QC (%) Methodology
Q3 51.7 75.2 24.8 56.2 MLNN
SOV 44.7 62.8 27.2 43.9
Q3 55 74.3 29.2 62.9 PSONN
SOV 52.3 74.8 32.0 54.9
Q3 67.9 95.4 32.7 77.1 FFANN
SOV 53.8 95.6 31.7 42.0
Figure 6.Accuracy of the prediction methodology
Figure 7.Calculated SOV of prediction methodology
From the above Table 5 and Figure 6&7, it is concluded that the proposed FFANN has improved prediction accuracy than the other conventional methodology.
Conclusion
A new topology called firefly trained neural fields which can able to tune NN automatically is designed for PSS prediction. In this work, a novel method firefly based ANN is implemented to identify SS of a protein. This proposed topology obtained is a promising accuracy over SS prediction than other topologies illustrated previously. This scheme automatically tunes the NN using optimization topology called firefly. The results obtained from the experimental setup have proven that this proposed topology can be utilized for predicting SS of any protein and hence, it would be more powerful in SS prediction research domain.
References
[1] Karpov, O. A., Fearnley, G. W., Smith, G. A., Kankanala, J., McPherson, M. J., Tomlinson, D. C., ... &Ponnambalam, S. (2015). Receptor tyrosine kinase structure and function in health and disease. AIMS Biophysics, 2(4), 476-502.2.
[2] Baker, D., &Sali, A. (2001). Protein structure prediction and structural genomics. Science, 294(5540), 93-96.
[3] Quan, L., Lv, Q., & Zhang, Y. (2016). STRUM: structure-based prediction of protein stability changes upon single-point mutation. Bioinformatics, 32(19), 2936-2946.
[4] Wang, S., Peng, J., Ma, J., &Xu, J. (2016). Protein secondary structure prediction using deep convolutional neural fields. Scientific reports, 6(1), 1-11.
[5] Whisstock, J. C., &Lesk, A. M. (2003). Prediction of protein function from protein sequence and structure. Quarterly reviews of biophysics, 36(3), 307.
[6] Radivojac, P., Clark, W. T., Oron, T. R., Schnoes, A. M., Wittkop, T., Sokolov, A., ...& Friedberg, I. (2013). A large-scale evaluation of computational protein function prediction. Nature methods, 10(3), 221-227.
[7] Eisenberg, D. (2003). The discovery of the α-helix and β-sheet, the principal structural features of proteins. Proceedings of the National Academy of Sciences, 100(20), 11207-11210.
[8] Yoo, P. D., Zhou, B. B., &Zomaya, A. Y. (2008). Machine learning techniques for protein secondary structure prediction: an overview and evaluation. Current Bioinformatics, 3(2), 74-86.
[9] Qian, N., &Sejnowski, T. J. (1988). Predicting the secondary structure of globular proteins using neural network models. Journal of molecular biology, 202(4), 865-884.
[10] Holley, L. H., &Karplus, M. (1989). Protein secondary structure prediction with a neural network. Proceedings of the National Academy of Sciences, 86(1), 152-156.
[11] Rost, B., & Sander, C. (1993). Prediction of protein secondary structure at better than 70% accuracy. Journal of molecular biology, 232(2), 584-599.
[12] Zvelebil, M. J., Barton, G. J., Taylor, W. R., & Sternberg, M. J. (1987). Prediction of protein secondary structure and active sites using the alignment of homologous sequences. Journal of molecular biology, 195(4), 957-961.
[13] Jones, D. T. (1999). Protein secondary structure prediction based on position-specific scoring matrices. Journal of molecular biology, 292(2), 195-202.
[14] Busia, A., &Jaitly, N. (2017). Next-step conditioned deep convolutional neural networks improve protein secondary structure prediction. arXiv preprint arXiv:1702.03865.
[15] Buchan, D. W., Minneci, F., Nugent, T. C., Bryson, K., & Jones, D. T. (2013).
Scalable web services for the PSIPRED Protein Analysis Workbench. Nucleic acids research, 41(W1), W349-W357.
[16] Mirabello, C., &Pollastri, G. (2013). Porter, PaleAle 4.0: high-accuracy prediction of protein secondary structure and relative solvent accessibility. Bioinformatics, 29(16), 2056-2058.
[17] Sun, T., Zhou, B., Lai, L., & Pei, J. (2017). Sequence-based prediction of protein protein interaction using a deep-learning algorithm. BMC bioinformatics, 18(1), 1-8.
[18] Gomes, J., Ramsundar, B., Feinberg, E. N., &Pande, V. S. (2017). Atomic convolutional networks for predicting protein-ligand binding affinity. arXiv preprint arXiv:1703.10603.
[19] Yang, X. S. (2009, October). Firefly algorithms for multimodal optimization.
In International symposium on stochastic algorithms (pp. 169-178). Springer, Berlin, Heidelberg.
[20] Gandomi, A. H., Yang, X. S., Talatahari, S., &Alavi, A. H. (2013). Firefly algorithm with chaos. Communications in Nonlinear Science and Numerical Simulation, 18(1), 89-98.
[21] Yang, X. S. (2010). Nature-inspired metaheuristic algorithms. Luniver press.
[22] Yang, X. S. (2010). Firefly algorithm, stochastic test functions and design optimisation. International journal of bio-inspired computation, 2(2), 78-84.
[23] Kryshtafovych, A., Barbato, A., Fidelis, K., Monastyrskyy, B., Schwede, T.,
&Tramontano, A. (2014). Assessment of the assessment: evaluation of the model quality estimates in CASP10. Proteins: Structure, Function, and Bioinformatics, 82, 112-126.
[24] Hua, S., & Sun, Z. (2001). A novel method of protein secondary structure prediction with high segment overlap measure: support vector machine approach. Journal of molecular biology, 308(2), 397-407.
[25] Liu, T., & Wang, Z. (2018). SOV_refine: A further refined definition of segment overlap score and its significance for protein structure similarity. Source code for biology and medicine, 13(1), 1-10.
