Protein Complex Inference enhanced by Text Mining

Protein complexes play a vital role in living organisms as they regulate and execute biological processes. As experimental methods of extracting protein complexes are fraught with difficulties, scientists look towards protein complex prediction. However, protein-protein interaction (PPI) data which are used to predict protein complexes are often noisy and incomplete. As published literature may hold a wealth of PPI data which goes unnoticed, this paper aims to enhance the prediction of protein complexes by text mining PPI data from literature abstracts. In this paper, we explore various rule-based methods of extracting PPI data from MEDLINE abstracts. Additionally, we show that the removal of non-hub proteins can reduce the impact of noisy PPI data on protein complex prediction and retrieve smaller and more accurate protein complexes which would otherwise be discarded by CMC. Moreover, we show that the selection of abstracts for augmentation is worth doing to overcome the incompleteness of PPI data. iii Acknowledgements I would like to express my deepest gratitude to my supervisor Professor Wong Limsoon for his patience and guidance during the course of this project. I would also like to thank Dr Liu Guimei for providing me with the source codes for CMC and evaluation using GO, Dr Rajesh Chowdhary for providing me the source codes for BN and his guidance in using the program, Su Zhan for his advice in the downloading of MEDLINE abstracts and Dr Upali Kohomban for his help in tagging the abstracts.