<p>Metaphors are important devices in academic discourse. Although several studies have examined the use of metaphors in academic discourse, their findings remain inconclusive due to the manual detection of metaphors and hence the very limited size of data used in their research. To address the issue, we proposed a method to analyse the use of lexical metaphors in a large-scale dataset of academic texts. To be specific, we used a pre-trained large language model to automatically identify and examine the use of lexical metaphors in more than forty-thousand abstracts across five disciplines. The findings showed that lexical metaphors were frequently used across disciplines. In addition, lexical metaphors were used more frequently in soft sciences than in hard sciences. Among the three types of metaphors, indirect metaphors were used most frequently. The results were explained from perspectives such as different roles that texts of different disciplines played for the dissemination of knowledge and the communication of scientific arguments and the functions of different metaphor types. Our study is the first large-scale attempt to extensively analyse the use of lexical metaphors with an automatic metaphor identification tool in academic discourse. Implications for academic writing instruction and future research are also discussed.</p>