NVIDIA GPU Accelerates Drug Discovery at Suikon Intelligence, Speeding Up Screening Process by Dozens of Times

SuiKun Intelligence leverages NVIDIA GPUs to significantly enhance the computational efficiency, accuracy, and model training speed of its AI4D™ new drug R&D platform, accelerating the entire drug molecule screening process by dozens of times and meeting the demands of processing massive datasets.

This case primarily utilizes the Tesla V100 GPU.

SuiKun Intelligence is a next-generation startup at the intersection of machine learning and biotechnology, dedicated to enhancing drug R&D efficiency and success rates by ten thousand-fold through its independently developed AI platform. The company has established collaborations with nearly 20 renowned domestic and international pharmaceutical companies, listed enterprises, Grade A tertiary hospitals, government agencies, and research institutes to jointly develop new drugs and AI models.

The success rate of new drug development is extremely low, and the R&D investment is substantial. The activity, drug-likeness, and developability of drug molecules are crucial for the successful completion of clinical trials and subsequent regulatory approval for market launch. Accurate prediction of these properties in the early stages of drug discovery can enhance the success rate of later-stage development.

The scale of big pharmaceutical data often reaches hundreds of thousands or even millions of records. The development and application of AI algorithms based on such ultra-large-scale datasets pose significant challenges to computational power. General-purpose CPU computing platforms suffer from slow processing speeds; typical deep learning model training requires tens of millions of iterations and parameter updates, consuming substantial time. The application of GPUs in the field of deep learning has accelerated the advancement of AI. Compared with CPU-based platforms, GPU acceleration reduces the time per iteration of deep learning model training from dozens of minutes to just a few seconds, significantly improving the efficiency of AI algorithm development, shortening development cycles, and reducing costs.

The AI4D™ online service platform developed by Suikun Intelligence utilizes NVIDIA Tesla V100 GPUs. Leveraging NVIDIA GPU-accelerated computing has significantly enhanced computational efficiency, precision, and model training speed for AI4D™. Compared to CPU-only systems, it increases throughput for real-world applications by 10-fold, substantially improving model training efficiency. The integration of GPUs has also increased double-precision performance by 10 times over earlier hardware, while significantly reducing data center costs by decreasing the number of required servers.

Massive-Scale Data Processing: During the literature information mining process, over one million literature texts need to be processed. With GPU acceleration, relationship extraction queries across the entire database can be completed within minutes, achieving a speed increase of more than 10-fold and significantly improving query efficiency.

Efficient Training of Deep Learning Models: The training speed of deep learning models is more than 100 times faster than that of CPUs, allowing for more iterations within the same timeframe. This significantly reduces time costs, effectively shortens the development cycle, and improves development efficiency.

Optimized Design of Biomacromolecules: The optimization space for biomacromolecules grows exponentially, and conventional optimization methods require substantial computational resources. Optimization algorithms developed on GPU platforms leverage GPU acceleration and multi-GPU parallelization strategies to reduce the time required for a single search and increase the number of searches per unit time. This enables the design and optimization of biomacromolecules to be completed within minutes, achieving an efficiency improvement of more than 40-fold.

The success of AI algorithms stems from their ability to extract deep insights from massive amounts of structured data. Suikun Intelligence leverages AI systems to integrate big pharmaceutical data, deeply empowering the entire drug R&D process. This encompasses a wide range of applications, including learning and predicting comprehensive landscapes of drug-target interactions, virtual high-throughput molecular screening, optimization of macromolecule yields, design of personalized cell therapies, and literature mining. The AI4D™ platform not only generates and screens target-specific molecules but also further predicts their drug-likeness and developability by drawing on the team’s extensive expertise in drug R&D. This approach helps focus on genuine drug candidates, thereby reducing late-stage R&D investments and improving the success rates for clinical trials and market approval.

On the AI4D™ platform, users can leverage AI for drug molecule screening with a low barrier to entry, eliminating the need to learn algorithms or programming. By delegating tedious, complex, and repetitive tasks to machines, users can focus their efforts on strategic research, significantly reducing experimental validation costs. Traditional drug molecule screening often takes six months to a year, sometimes yielding no results at all. Leveraging NVIDIA’s computing platform to accelerate machine learning increases the throughput of the entire computational screening process by tens of times, meeting the demands of massive data processing, drastically cutting time costs, and further accelerating new drug development.

“Machine learning is a groundbreaking tool in the field of drug discovery, with vast application prospects in the biomedical sector. Although machine learning does not demand excessive computational power, fitting biological mechanisms’ high complexity with machine learning models requires more robust computing capabilities. The NVIDIA AI computing platform can further advance the integration of machines and life,” said Zeng Hainian, CEO of Suikun Intelligence.