Extraction of gene/protein names involved in each stage of spermatogenesis based on literature mining

Spermatogenesis is an important bioprocess in the lifetime of male mammalians, which has deep effect on mammal's reproduction. Abnormal spermatogenesis is a major cause of male infertility, however treatments for this are limited. Characterizing the genes/proteins involved in spermatogenesis is fundamental to understand the mechanisms underlying this biological process and to develop treatments for the problems in spermatogenesis. However, most crucial information of spermatogenesis-related genes/proteins scatters in vast amount of research articles, so manually curation of these genes/proteins could be a time-consuming task. In this paper, a novel strategy is proposed to automatically extract the names of spermatogenesis-related genes/proteins, which function in different stages of spermatogenesis based on literature mining. Firstly, it compares three different algorithms performance on different terms and applys an SVM classifier trained with a manually prepared dataset to classify spermatogenesis-related texts into three classes in accordance with the three stages of spermatogenesis. Then, integrating expert knowledge and grammar rules, it recongnizes and extracts the gene/protein names of each spermatogenesis stage with high confidence. Finally, a manually curation test dataset is used to test the performance of this strategy, and the strategy gets an accuracy of 71.9%, which verifys the reliability of proposed method and proves the value of application.