Support Vector Machine – a model for QSAR still robust in the era of Deep Learning
The new QSAR framework in Flare makes generating and using 3D-QSAR descriptors to build models with good predictivity and generalizability ...
There is a revolution happening in high performance computing. Calculations that once took hours or days to run can now be done in just minutes. The secret? Graphics Processing Units (GPUs), commonly called “graphics cards”. Graphics cards are no longer limited to just moving triangles around on screen: they are now phenomenally powerful parallel computers in their own right. They can only do a limited subset of tasks, but do them at lightning speed.
Cresset is working to bring this exciting revolution to all our customers. We have have a collaboration with the high performance computing group at the University of Bristol, UK that will implement new GPU based algorithms within the core of our field technology. Our collaboration takes the form of a Knowledge Transfer Partnership (KTP) scheme from the UK government which provides a way to transfer knowledge held in academia out into industry. As part of this we have recruited Simon Krige, a GPU programming expert, to the Cresset team. Simon is working closely with Prof. Simon McIntosh-Smith at Bristol whose group is one of the world leaders in high-performance parallel computing, and is helping to bring their expertise into Cresset.
Simon (Krige) has made great advances in the few months that he has been with us. He has rewritten large sections of the core field similarity algorithm in openCL, a framework that will execute on either regular CPUs or GPUs. The beauty of this system is that the code runs wherever it can go fastest, on your CPU if you do not have a fast GPU or on the GPU if you do. In this way speed increases should be available for desktop as well as cluster based applications.
So far the numbers are impressive with speed increases of 40x being observed in a virtual screening environment on standard graphics hardware. Simon is eager to point out that this may not be maintained when we are in a desktop environment where the workload is more sporadic. However there is plenty of optimization of the code still to do and so speeds could go up as well as down.
As part of our development schedule we are looking for partners to start alpha testing in the next couple of months. If you are interested in collaborating with us on this please email us.