Custom Computing for Phylogenetics: Exploring the Solution Space for UPGMA

There is great interest in the use of custom computing machines (CCMs) to accelerate the performance of existing, highly optimized software algorithms in Bioinformatics. But, like many applications for CCMs in the sciences, there are many architecture and experimentation “data points” to be obtained before a solid understanding can be obtained as to the issues of migrating specialized scientific computing software applications to high-performance CCM platforms. This paper presents one such data point, namely, the design and comparative analysis of a custom computing application architecture for the UPGMA algorithm implemented on an FPGA-based platform. We present the UPGMA problem domain and explore design issues encountered in the transition from software algorithm executing on a standard PC platform to a hybrid custom compute “server” running a VLSI logic-based architecture for the algorithm. Using the low-cost Annapolis WildCard® platform, we discuss the CCM development method, and present an architecture created that takes into consideration the processing limitations of the platform. Given this implementation, we present results of experiments on system performance, as measured and compared against that of the benchmark UPGMA algorithm written in C, executing on a singleprocessor Pentium® PC.