Textomics: A Dataset for Genomics Data Summary Generation

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 4878 - 4891 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Textomics: A Dataset for Genomics Data Summary Generation Mu-Chun Wang1,∗, Zixuan Liu2,∗, Sheng Wang2 1University of Science and Technology of China 2Paul G. Allen Scholl of Computer Science and Engineering, University of Washington amoswang2000@mail.ustc.edu.cn {zucksliu, swang}@cs.washington.edu Abstract Summarizing biomedical discovery from ge- nomics data using natural languages is an essential step in biomedical research but is mostly done manually. Here, we introduce Textomics, a novel dataset of genomics data description, which contains 22,273 pairs of genomics data matrices and their summaries. Each summary is written by the researchers who generated the data and associated with a scientific paper. Based on this dataset, we study two novel tasks: generating textual sum- mary from a genomics data matrix and vice versa. Inspired by the successful applications of k nearest neighbors in modeling genomics data, we propose a kNN-Vec2Text model to address these tasks and observe substantial im- provement on our dataset. We further illustrate how Textomics can be used to advance other applications, including evaluating scientific pa- per embeddings and generating masked tem- plates for scientific paper understanding. Tex- tomics serves as the first benchmark for gen- erating textual summaries for genomics data and we envision it will be broadly applied to other biomedical and natural language process- ing applications.1 1 Introduction Modern genomics research has become increas- ingly automated through being roughly divided into three sequential steps: next-generation sequenc- ing technology produces a massive amount of ge- nomics data, which are in turn processed by bioin- formatics tools to identify key variants and genes, and, ultimately, analyzed by biologists to summa- rize the discovery (Goodwin et al., 2016; Kanehisa and Bork, 2003). In contrast to the first two steps that have been automated by new technologies and ∗Equal Contribution 1The link to access our code: https://github.com/ amos814/Textomics software, the last step of summarizing discovery is still largely performed manually, substantially slowing down the progress of scientific discovery (Hwang et al., 2018). A plausible solution is to automatically summarize the discovery from ge- nomics data using neural text generation, which has been successfully applied to radiology report generation (Wang et al., 2021; Yuan et al., 2019) and clinical notes generation (Melamud and Shiv- ade, 2019; Lee, 2018; Miura et al., 2021). In this paper, we study this novel task of gen- erating sentences to summarize a genomics data matrix. Several excisting approaches demonstrate encouraging results in generating short phrases to describe functions of a set of genes (Wang et al., 2018; Zhang et al., 2020; Kramer et al., 2014). However, our task is fundamentally different from these: the input of our task is a matrix that contains tens of thousands of genes, which could be noisier than a set of selected genes; the outputs of our task are sentences instead of short phrases or controlled vocabularies. To study this task, we curate a novel …

Content truncated to 500 words