Transforming computational drug design with Google Cloud

Computationally predicting an effective drug molecule is an extremely challenging problem with over fifty years of significant research. Until recently, the limitation of compute power had prevented pursuing any truly accurate solutions. Now that we’re able to make accurate predictions, additional problems arise in storing and querying predictions made against nearly infinite possible drug molecules.

By utilizing many of the tools of Google Cloud, the compute and data storage limitations of computational drug design are finally becoming things of the past. Google Cloud transforms drug discovery, using massive-scale GPU compute on Google Cloud to drive a physics-based active learning approach. Listen to a discussion of the many ways Google Cloud addresses storing and querying vast data, including the extensive use of BigQuery and custom high-memory instances.

Speaker: Pat Lorton

Watch more:

#GoogleCloudNext

CMP114
product: Compute Engine, Migrate for Compute Engine;

re_ty: Publish; product: Cloud - Compute - Migrate for Compute Engine; fullname: Pat Lorton; event: Google Cloud Next 2020;

[nguyenngocly]

You can have inside-out neural networks with fixed dot prroducts and adjustable (parametric) activation functions. You can use fast transforms like the FFT for the fixed dot products. Hence Fast Transform (fixed filter bank) neural networks. The dot products are statistical in nature (summary measures) which helps a lot.