Non-linear correlation of content and metadata information extracted from biomedical article datasets

Area of Science:

• Biomedical Informatics
• Machine Learning
• Data Science

Background:

• Biomedical literature databases are crucial for accessing current scientific knowledge.
• Organizing these databases and extracting novel insights necessitates efficient machine learning methods.
• Many machine learning methods require documents to be represented as large multivariate datasets.

Purpose of the Study:

• To investigate Non-Linear Canonical Correlation Analysis (NLCCA) for integrating information from different document representations.
• To address the challenge of combining voluminous information from various sources into a concise, representative dataset.
• To describe documents using a single, consolidated dataset.

Main Methods:

• Utilized Non-Linear Canonical Correlation Analysis (NLCCA), a multivariate statistical approach.
• Exploited correlations among variables from different document representations (text words, MeSH, GO terms).
• Experimented with document datasets represented by text words, Medical Subject Headings (MeSH), and Gene Ontology (GO) terms.

Main Results:

• Demonstrated the effectiveness of NLCCA in combining information from diverse biomedical data sources.
• Showcased NLCCA's ability to create a concise dataset that efficiently represents a corpus of documents.
• Validated the approach using document datasets represented by text words, MeSH, and GO terms.

Conclusions:

• NLCCA is an effective method for integrating heterogeneous biomedical data sources.
• The approach facilitates the creation of concise, informative datasets for improved knowledge discovery.
• NLCCA shows promise for enhancing the organization and extraction of knowledge from biomedical literature.