High performance-oriented computer aided drug design approaches in the exascale era

Area of Science:

• Computational science
• Drug discovery
• High-performance computing (HPC)

Background:

• The advent of exascale supercomputing, exemplified by the Frontier system, marks a significant advancement in computational power.
• Exascale computing presents opportunities to revolutionize fields like computer-aided drug design (CADD).
• Scaling existing CADD approaches for exascale architectures necessitates novel algorithmic and software solutions.

Purpose of the Study:

• To explore the application of physics-based and machine learning (ML)-aided techniques for small molecule binder design on exascale systems.
• To review HPC-oriented large-scale CADD applications from the past three years that utilized pre-exascale and exascale supercomputers.
• To identify the potential and current limitations of leveraging exascale computing for advanced drug design.

Main Methods:

• Review of recent (3-year) HPC-oriented large-scale applications in CADD.
• Analysis of physics-based methods adapted for parallel computer architectures.
• Evaluation of machine learning (ML) approaches, including generative models and ML-aided physics-based methods.

Main Results:

• Exascale computing can enable the training of highly predictive generative models for novel ligand design, given sufficient data.
• Accurate ML-aided physics-based methods on exascale systems show promise for enhancing structure-based drug design success rates.
• Current methodological advancements are still needed for the routine, large-scale application of these rigorous CADD approaches.

Conclusions:

• Exascale computing holds transformative potential for computer-aided drug design.
• Integrating physics-based and ML methods on exascale platforms can accelerate the discovery of small molecule binders.
• Further methodological development is crucial to fully realize the benefits of exascale-driven drug design.