Computational approaches for predicting biomedical research collaborations.

Area of Science:

• Biomedical informatics
• Computational biology
• Bibliometrics

Background:

• Biomedical research collaborations are crucial for high-impact work.
• Existing collaboration prediction methods primarily use network topology, neglecting publication semantics.
• There is a need for advanced computational approaches to predict biomedical collaborations.

Purpose of the Study:

• To propose and evaluate supervised machine learning models for predicting biomedical research collaborations.
• To investigate the utility of semantic features derived from author research interests alongside network topological features.
• To identify the most effective features and models for collaboration prediction.

Main Methods:

• Extracted semantic features from author research interest profiles, including citation and abstract similarity.
• Incorporated author network topological features into prediction models.
• Applied and compared four supervised machine learning models: naïve Bayes, naïve Bayes multinomial, Support Vector Machines (SVMs), and logistic regression.

Main Results:

• Semantic features related to research interest, such as similarity in out-citing citations and abstracts, were highly informative.
• Logistic regression demonstrated the best performance, achieving an ROC range of 0.766 to 0.980 across different datasets.
• The proposed approach is computationally efficient, scalable, and simple to implement.

Conclusions:

• Author research interests and productivity, when analyzed computationally, significantly improve biomedical collaboration prediction.
• Logistic regression is a highly effective model for this task.
• This study provides a novel, efficient, and scalable method for predicting scientific collaborations, with available datasets for reproducibility.