Generic information can retrieve known biological associations: implications for biomedical knowledge discovery

Area of Science:

• Biomedical Informatics
• Computational Biology
• Text Mining

Background:

• Weighted semantic networks from text mining aid in discovering biological associations.
• Biomedical text mining systems recognize numerous specific and generic concepts.
• Generic concepts can cause information overload and spurious links in retrieval.

Purpose of the Study:

• To optimize protein-protein interaction (PPI) retrieval by filtering generic concepts from semantic networks.
• To evaluate the impact of filtering generic vs. specific information on network retrieval performance.
• To define metrics for concept specificity to guide network filtering.

Main Methods:

• Developed metrics to quantify concept specificity based on network properties.
• Systematically filtered generic and specific information from weighted semantic networks.
• Monitored retrieval performance of known protein-protein interactions during filtering.

Main Results:

• Filtering generic or specific information caused a sudden drop in network performance beyond a critical threshold.
• Networks composed solely of generic concepts showed retrieval performance comparable to unfiltered networks.
• Individual generic concepts, like "binding," effectively support PPI retrieval.

Conclusions:

• Generic concepts are vital for effective information retrieval in semantic networks.
• Removing generic concepts negatively impacts retrieval performance.
• Generic concepts play a significant role in anticipating and retrieving biological associations.