FastBit: Digging through
databases faster
(page 5 of 5)
Other applications push FastBit into medicine. For example, Jochen Schlosser, a doctoral student in bioinformatics at the University of Hamburg, Germany, uses FastBit to improve the discovery of new drugs. If a particular protein plays some role in a disease, for example, scientists search for a molecule that blocks said protein from doing its disease-related task. A large pharmaceutical company, though, might have a million or more compounds in a potential-drug library, and testing them all for a structural fit with every disease target takes immense amounts of computation.
So Schlosser and his colleagues developed TrixX BMI. This software generates information about every compound – so-called TrixX descriptors – that get stored in a compressed bitmap index. “You build an abstract of the compound and target protein to decide on an abstract level if the compound is a match or not,” Schlosser says. That abstract description encodes shape, molecular interactions and directional constraints in roughly 100 dimensions.
Instead of looking at every aspect of every compound in a library, TrixX BMI can eliminate compounds that have features that make them unlikely to dock with a specific protein. Using FastBit, the query phase of TrixX BMI runs 5 to 10 times faster than fully scanning the raw data.
“On average, the query phase is about one-fifth of the total runtime,” Schlosser says. Once FastBit finds compounds that look likely to dock with a specific protein, then Schlosser and his colleagues turn to further – and more computationally expensive – simulations.
FastBit’s future
FastBit’s capabilities make it useful for a wide range of problems. “Our primary goal at this point,” Wu says, “is trying to find new applications that could motivate our work. For example, we are currently working with some fusion-simulation groups, doing analysis of their data and dealing with interesting geometries.”
Likewise, some mechanical fine-tuning could lie ahead. FastBit grew out of research and it lacks some user amenities. “It was not exactly geared for production software,” Wu says. “If we get a chance, we might re-implement some of FastBit.”
Meanwhile, researchers keep putting FastBit to work, indexing and searching a wide range of high-dimensional and expansive databases. And the results will continue to teach us more about our world.
