Virk: an active learning-based system for bootstrapping knowledge base development in the neurosciences.

Area of Science:

• Neuroscience
• Bioinformatics
• Computational Biology

Background:

• Manual curation of scientific databases is becoming unfeasible due to the increasing volume of published literature.
• Machine learning (ML) offers automated solutions, but requires extensive manual annotation for training.
• Existing active learning (AL) methods are insufficient for efficient scientific database curation.

Purpose of the Study:

• To develop an automated system (Virk) for scientific database curation that simultaneously trains a classification model and identifies relevant documents.
• To improve the efficiency and scalability of knowledge base maintenance.

Main Methods:

• Implemented an active learning (AL) system named Virk.
• Utilized a support vector machine (SVM) classifier with features extracted from neuroscience publications.
• Integrated model training and document identification within the AL framework.

Main Results:

• Virk increased the size of the Neuron Registry knowledge base by 90% in 3 months.
• Achieved in 3 months what would take 1-2 years using standard biocuration methods.
• Demonstrated superior performance compared to other AL sampling approaches.

Conclusions:

• The Virk system offers a highly efficient and scalable solution for scientific database curation.
• Automated curation using Virk significantly accelerates the growth of neuroscience knowledge bases.
• This approach addresses the limitations of manual curation and standard active learning in the context of big data in science.