Can the vector space model be used to identify biological entity activities?

Area of Science:

• Bioinformatics
• Computational Biology
• Life Sciences

Background:

• Biological systems are often modeled as interaction networks, with many interactions remaining undiscovered.
• Existing knowledge bases integrate known biological interactions, but discovering novel ones is challenging.
• This work addresses the need for methods to predict previously unknown biological interactions.

Purpose of the Study:

• To develop and validate a model for predicting unknown biological interactions from textual collections.
• To leverage the vector space model (VSM) with a transitive closure approach for enhanced information retrieval.
• To construct a biological interaction network and identify novel, high-ranking interactions.

Main Methods:

• Utilized the vector space model (VSM), a standard information retrieval technique.
• Extended VSM capabilities by incorporating a transitive closure approach.
• Applied the model to a collection of patent claims (1976-2005) to identify and predict interactions.

Main Results:

• The model identified 1,027 known interactions and predicted 3,195 new interactions from 266,528 possibilities.
• Temporal analysis confirmed predicted interactions with later-issued patent claims and external literature.
• The top predicted interaction involved the adrenaline neurotransmitter and the androgen receptor gene, later supported by research.

Conclusions:

• The VSM combined with transitive closure effectively identifies and ranks novel biological interactions from text.
• This approach facilitates literature-based discovery, highlighting relevant new interactions even in sparse data.
• The developed method offers an efficient way to extract and prioritize potential breakthroughs in life sciences.