-
AlgoPha-Shape
AlgoPha-Shape is a highly efficient molecular shape matching algorithm using Gaussian spheres to represent atoms of molecules. The GPU implementation of this algorithm speed up computation by two orders of magnitude. The overall procedure can be shown as follows:
As the GPU accelerated computation is so fast that the IO of loading conformation into CPU memory and then send the conformation to GPU memory becomes the bottleneck. To solve this problem, we have developed a very efficient data compression method to store xyz coordinates and use memory map file and SIMD speed up data conversion. The overall workflow is shown as follow:
High Throughput: with all the tricks as shown above, we have achieve speed of 1.3 million shape comparison per second with a single K80 GPU.
-
AlgoPha-Conf
This is an extremely efficient conformation search algorithm. It generates conformations for several hundreds of drug-like molecules and each molecule with about 100 conformations. The algorithm is a completely new implementation of the original CAESAR algorithm with significant enhancements, including the 64-bit encoding of molecules and fragments and efficient reused of fragment conformations of molecular building blocks. An extremely efficient 3D conformation presentation method has been invented which can reduce the conformation storage by 2-3 orders of magnitudes.
-
AlgoPha-CM
AlgoPha-CM is a charge model for drug-like molecules. It assign atomic point charges according to the atomic type and its surrounding chemical environment. The charge model is trained with massive high quality Quantum Chemistry calculations so that the assigned atomic charges can reproduce the dipole moments of the molecules. This model, after training, can be applied to any organic molecules without doing any Quantum Chemistry calculations, therefore it is extremely fast.
-
AlgoPha-VB (VB2000)
AlgoPha-VB is a modern Valence Bond calculation package. It is derived from the free-distributed VB2000. AlgoPha-VB is based on Algebrant algorithm. The original VB2000 package has been revised and expanded for much large VB calculations. It can be used to run VB calculations with up to 20 electrons in each VB group on a home desktop. Calculations for 22 electron VB group have been tested on computers with very large memory (~100GB).
-
CXAI-Profiler
A customer profiling system to measure customer’s intention for various business activities, such as the expectation for revisit a store, customer’s business values, the loyalty measurement.
-
CXAI-Recommender
A recommendation system to make recommendation to loyal customers based on their historical behavior.
-
CXAI-Playground
A cloud based server for students to learn Python programming and Machine Learning. Class materials and projects are provided for registered students.