Autoparty: Machine Learning-Guided Visual Inspection of Molecular Docking Results.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Human inspection is vital but slow in virtual drug screening.
• Current methods lack standardization and consistency.
• Scalability is a major challenge due to the vast number of molecules.

Purpose of the Study:

• To introduce Autoparty, a containerized tool to accelerate human-in-the-loop drug discovery.
• To enable efficient training of models that learn from human expertise.
• To standardize annotation recording and create a persistent database.

Main Methods:

• Leveraging on-premises active learning for drug discovery.
• Employing uncertainty quantification metrics for informative user queries.
• Developing a containerized tool for streamlined workflow.

Main Results:

• Autoparty facilitated human-in-the-loop model training.
• Uncertainty quantification reduced the need for extensive human labels.
• A 40% increase in hit rate was observed in a case study.

Conclusions:

• Autoparty effectively accelerates virtual drug screening.
• The tool enhances model training by extrapolating human intuition.
• Standardized annotations and a local database support downstream applications.