The future of machine learning for small-molecule drug discovery will be driven by data.

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Biotechnology

Background:

• Machine learning (ML) integration in small-molecule therapeutics development is widely anticipated to accelerate drug discovery.
• Despite advancements in ML algorithms and architectures, substantial improvements in therapeutic development outcomes have been limited.

Purpose of the Study:

• To propose that enhanced focus on data quality for training and benchmarking ML models is crucial for future progress in drug discovery.
• To explore research avenues and strategies for addressing data-related challenges in ML-driven drug development.

Main Methods:

• This perspective synthesizes current trends and challenges in applying machine learning to drug discovery.
• It emphasizes a shift from algorithmic complexity to data-centric approaches for model development and validation.

Main Results:

• Current ML applications in drug discovery show diminishing returns with algorithm-only improvements.
• A data-centric approach is identified as a more promising strategy for enhancing model performance and reliability.

Conclusions:

• Prioritizing high-quality, well-curated datasets for training and benchmarking is essential for realizing the full potential of machine learning in small-molecule therapeutics.
• Future research should focus on data generation, standardization, and innovative methods to overcome data limitations in ML-driven drug discovery.