Scaling of AI-based protein design
CS-21366
HigherEd
TUM Technische Universität München
We are the chair for biomolecular nanotechnology at the Technical University of Munich. We try to develop functional nanostructures and molecular machines with applications in research, medicine or materials science and use methods such as deoxyribonucleic acid (DNA) -origami Technology or protein design.
The Challenge
With the advent of AI-based development methods for the design of proteins, the hardware requirements also grow. For the high -scaling of protein design algorithms, high -performance GPUs are required, which are very expensive to buy and are therefore not suitable for our pilot experiments.
The Solution
With the Google Cloud we had quick and cost -effective access to NVIDIA A100 GPUs with which we received the required computing power. With the simple cloud console and Vertex AI, integrating our codes was not a problem.
The Results
With the help of the Cloud GPUS, we were able to successfully scale our design algorithms to up to 1000 amino acids, make the necessary debugging and the biggest de novo designed proteins.
The Highlight
The greatest benefit was definitely the flexibility of the Google Cloud and the strong support also by the experts who were provided to us.