Antibody Watch: Text mining antibody specificity from the literature

Area of Science:

• Biotechnology
• Bioinformatics
• Scientific Literature Analysis

Background:

• Antibodies are crucial research tools, but their lack of specificity can lead to unreliable results.
• Ensuring antibody specificity is challenging due to the vast number of available reagents.

Purpose of the Study:

• To investigate the feasibility of automatically identifying and alerting researchers about problematic antibodies using literature mining.
• To construct a knowledge base of antibody specificity issues from scientific publications.

Main Methods:

• Developed a deep neural network system to process over two thousand research articles.
• Implemented a two-task approach: identifying specificity snippets and linking them to specific antibodies using RRIDs.

Main Results:

• The system achieved high performance in classifying problematic antibody statements (0.925 weighted F1-score).
• Accurate linking of specificity issues to antibodies was demonstrated (0.962 accuracy).
• The combined task achieved a 0.914 weighted F1-score, proving the system's effectiveness.

Conclusions:

• Automated text mining is a feasible approach for building a reliable knowledge base of antibody specificity concerns.
• This system can enhance research reproducibility by alerting scientists to potentially unreliable antibodies.