Bagi memahami corak-corak lipatan protein dan fungsinya, pengetahuan berkaitan kelas struktur protein merupakan satu elemen yang tidak harus diabaikan. Dengan itu, kaedah meramal kelas struktur protein yang berkesan harus diterokai. Banyak algoritma telah dicadangkan dalam tempoh tiga dekad yang lalu. Namun, hanya beberapa kaedah yang dapat menunjukkan keputusan peramalan yang memuaskan, dan yang paling ketara adalah meramal dengan menggunakan kaedah penghitungan jarak kompleksiti dan pengelas jiran terdekat. Walau bagaimanapun, pendekatan ini memerlukan tempoh peramalan yang panjang akibat penggunaan urutan asid amino yang lengkap dalam proses peramalan. Justeru, satu kaedah baru dibentangkan, iaitu ramalan kelas struktur protein menggunakan peranti FPGA , dengan harapan untuk memperbaiki kelajuan ramalan. Kaedah ini berasaskan penghitungan jarak kompleksiti dan pengelas jiran terdekat. Jarak antara sepasang urutan protein dibina dengan penghitungan berdasarkan kompleksitinya dan bukan dengan pengekstrakan ciri-cirinya. Pengelas jiran terdekat digunakan sebagai alat ramalan dan sistem ini diimplimentasikan dalam Altera Cyclone II FPGA (EP2C35) dengan menggunakan bahasa pengatucaraan verilog. Keputusan daripada simulasi, pendekatan perisian dan bantuan peranti membuktikan bahawa kaedah ini dapat berfungsi sebagai peramal kelas struktur protein dan memaparkan keputusan 40 kali ganda lebih laju berbanding dengan pendekatan perisian tanpa peranti pemecut.
_______________________________________________________________________________________________________
In order to understand protein folding patterns and its function, adequate knowledge of protein structural class is an element that should not be ignored. Thus, it is important that effective methods for protein structural class prediction are developed. For the past three decades, many algorithms had been proposed but only few of them performed well in terms of accuracy. The most significan tone was the nearest neighbour complexity distance measure (NN-CDM) method. However, this approach suffered from high computational load due to utilization of complete amino acid sequence in prediction. To this end, a new effort is presented: prediction of protein structural class using a Field-Programmable Gate Array (FPGA) based hardware accelerator, hoping to improve prediction of protein structural class in terms of its speed. It is a nearest neighbour classifier with complexity-based distance measure. Distance between pair of protein sequences is evaluated by a complexity based measure o fraw symbol sequences rather than extracting features from them. And as predictive engine, nearest neighbour classifier is adopted and is implemented into Altera Cyclone II FPGA (EP2C35) using verilog hardware description language (HDL). Results obtained from simulation, software approach and hardware implementation proved that this approach can function as a protein structural class predictor and it is 40 times faster compared to software approach, without any hardware accelerated.
Prediction of protein structural class using a fpga-based hardware accelerator / Yee Chau Jinn