Volcano Engine helps Shenzhen Technology release the industry's first 3D molecular pre-training model Uni-Mol

The continuous emergence of new drugs has improved the quality of human life and average life span to a certain extent. In the field of drug design, since there are tens of millions of combinations of drug molecules and diverse structures, the chemical space of drug candidates has become very large.

#How can we speed up drug design? Shenzhen Technology, established in 2018, is committed to using artificial intelligence and molecular simulation algorithms, combined with advanced computing methods, to solve this important scientific problem.

Challenges of Molecular Simulation Algorithms

Shenzhen Technology is a practitioner of the "AI for Science" scientific research paradigm and is the most advanced model of human civilization. Basic biomedicine, energy, materials and information science and engineering research create a new generation of micro-scale industrial design and simulation platforms.

#This high-tech enterprise has strong scientific research and industrial implementation capabilities. For example, in the field of medicine, Shenzhen Technology combines free energy perturbation theory, molecular dynamics, enhanced sampling algorithms with high-performance computing through Hermite Uni-FEP, Uni-Fold, RiD and other modules to accurately predict protein structure and conformational changes. , and efficiently evaluate the binding free energy of proteins and ligands with chemical precision, provide efficient and accurate theoretical guidance for drug developers, and improve the efficiency of drug design and optimization.

Since molecular simulation algorithms are linearly dependent on computing power and other resources, Shenzhen Technology has encountered the following challenges in terms of underlying resource scheduling and algorithm optimization:

1. The number of computing tasks is large and fluctuates greatly, requiring the underlying platform to provide flexible and efficient elastic scheduling capabilities;
2. Large model, large memory training;
3. Reading massive unstructured file data;

Shenzhen Technology chooses to cooperate with Volcano Engine to obtain efficient, reliable and flexible high-performance computing services.

The industry’s first three-dimensional molecular pre-training model

The Volcano Engine provides Shenzhen Technology with a scientific computing solution based on GPU acceleration, which can Automatically elastically expand capacity based on business traffic, automatically collect the health status of each node, and formulate corresponding self-healing strategies to achieve intelligent operation and maintenance. For example, when a node is unavailable, this solution can replace and reschedule the containers on the node to ensure the smooth and healthy operation of the business.

The scientific computing solution relies on the cloud-based products of Volcano Engine to provide elastically scalable accelerated computing server GPUs with high performance A file service cluster and ultra-high concurrency throughput parallel file storage, coupled with a machine learning platform specially designed for high-performance computing scenarios such as AI and scientific computing, empower Shenzhen Technology to focus on core business development.

The relevant person in charge of Shenzhen Technology said: "After using scientific computing solutions, we can focus on building the application itself without directly purchasing and managing infrastructure such as underlying servers. Effectively reducing IT maintenance costs."

The Volcano Engine scientific computing solution effectively guarantees the research and development progress of Shenzhen Technology. Recently, Shenshi Technology released the first three-dimensional molecular pre-training model Uni-Mol (https://github.com/dptech-corp/Uni-Mol).

##Uni-Mol framework diagram

Uni-Mol directly connects the molecules Three-dimensional structures are used as model inputs rather than one-dimensional sequences or two-dimensional graph structures. Representation learning based on three-dimensional information allows Uni-Mol to surpass SOTA (state of the art) in almost all downstream tasks related to drug molecules and protein pockets, and also allows Uni-Mol to directly complete molecular conformation Generation, protein-ligand binding conformation prediction and other tasks related to 3D conformation generation, and surpass existing solutions.

In the future, Shenzhen Technology will also launch a series of protein free energy and absolute free energy calculation functions based on the Volcano Engine platform, bringing innovative solutions to the industry. Advanced productivity tools accelerate the development of pharmaceutical and other design designs.

The above is the detailed content of Volcano Engine helps Shenzhen Technology release the industry's first 3D molecular pre-training model Uni-Mol. For more information, please follow other related articles on the PHP Chinese website!