The middle path may hold the best shot for protein structure prediction. Scientists at the Indian Institute of Technology in New Delhi say the solution to the protein-folding problem is unlikely to spring from extraordinary trust in bioinformatics alone or entirely through the basic laws of physics.
Professor Bhyravabhotla Jayaram and his colleagues at the IIT chemistry department have combined bioinformatics and molecular dynamics to develop what they claim is a shortcut to protein folding. The technique involves generating a very large set of possible 3D protein structures from the sequence of amino acids and using a set of biophysical filters to eliminate unlikely structures.
The filters are a collection of rules and algorithms — some extracted from known protein structures through conventional bioinformatics data-mining, and others that use those laws of physics — to rule out 3D structures disallowed by nature. “These filters shrink the search space for likely structures, so when we finally get down to applying molecular dynamics we have a small number of structures to investigate,” says Achintya Das, a computational chemist at IIT.
But even that narrowed set of structures for large proteins — containing 500 or more amino acids — could contain several thousand, or even tens of thousands, of possibilities. The only way to deal with such a volume would be number-crunching supercomputers, but the IIT team is unlikely to complain about hardware any time soon.
India’s department of biotechnology has invested $200,000 at IIT to set up the country’s first exclusive bioinformatics supercomputing facility, with a matching grant from Sun Microsystems. The 70-processor Sun cluster, added to existing Pentium clusters and Silicon Graphics machines, gives the facility a computing power of 150 gigaflops.
The new technique, running on four processors, could fold a test protein containing 36 amino acids in 14 days, Jayaram said at the facility’s inauguration. It had taken a team at University of California, San Francisco, nearly two months and 256 processors to fold the same protein through pure molecular dynamics.
The IIT methodology has already attracted industry interest. “We’d like to see whether this works on some hypothetical proteins of Mycobacterium tuberculosis that we’re looking at as potential drug targets,” says Indira Ghosh, a senior scientist at AstraZeneca’s Research and Development Center in Bangalore.
— Ganapati Mudur