Institute of Information Science, Academia Sinica

Emerging evidence for regulation by non-coding RNAs—untranslated transcripts often reside in the intergenic regions (or the “dark side”) of human genome—in cancer development and therapeutic resistance opens up new avenues for cancer diagnosis and treatment. Taking advantage of several large-scale cancer genome studies on hundreds of cell lines and thousands of patients, we reverse-engineered regulatory networks to study their potential for pathophysiological effects in isolation and within integrative frameworks. In this presentation, I will introduce several machine learning-based approaches developed in our lab, in conjunction with in vitro and in vivo experimental data, to demonstrate how these non-coding RNAs confer cancer-relevant functions. The complete list of approaches and their brief descriptions are as follows. Hermes (Sumazin et al., Cell 2011): Through an analysis of large-scale molecular profiling experiments in glioblastoma multiforme, we uncovered a post-transcriptional regulatory layer comprising hundreds of thousands of miRNA-mediated interactions (or ceRNA interactions). We showed that these interactions provide channels for the propagation of genetic alterations, pointing to the origins of unexplained regulation. Cupid (Chiu et al., Genome Res. 2015): Accurate miRNA-target prediction is needed to improve our understanding of miRNA-mediated regulation, but current methods are notoriously inaccurate. We developed Cupid to first score miRNA binding sites using sequence-based binding-site predictions, then it predicts interactions using the scored sites, and finally, it predicts functional, context-specific interactions using evidence for miRNA targeting. Longhorn (Chiu et al., Cell Rep. 2018): We inferred lncRNA-interaction networks, including inferred transcriptional and post-transcriptional pan-cancer interactions, to provide a resource for studying lncRNA regulation and the effects of lncRNAs on tumor etiology in a multitude of cancer contexts. Longhorn is a part of TCGA’s Pan-Cancer Atlas efforts. It is being applied to another international project with a collection of 300 human sequencing samples using strand-specific polyA, total RNA, and small RNA libraries. Bighorn (Chiu et al., in preparation): We inferred DNA binding motifs and transcriptional targets of lncRNAs in sixteen tumor types. In addition, we identified an lncRNA that regulates DNA damage response and repair pathways and is predicted to improve the effectiveness of X-ray therapy. This lncRNA is also predicted to regulate the transcription of DICER1, an important miRNA biogenesis regulator.

BEDUCATION 2014 Ph.D. in Biomedical Informatics (bioinformatics track), Columbia University, New York, NY 2012 M.Phil. in Biomedical Informatics (bioinformatics track), Columbia University, New York, NY 2011 M.A. in Biomedical Informatics (bioinformatics track), Columbia University, New York, NY 2004 M.S. in Computer Science and Information Engineering, National Taiwan University, Taiwan 2002 B.S. in Computer Science and Information Engineering, National Central University, Taiwan (GPA 3.91—ranked 1st among 75 graduates) ACADEMIC APPOINTMENTS 2019–current Staff Scientist, Pediatrics, Baylor College of Medicine, Houston, TX 2014–2018 Research Associate, Pediatrics, Baylor College of Medicine, Houston, TX 2010–2013 Graduate Research Assistant, Systems Biology, Columbia University, New York, NY 2004–2009 Research Assistant, Institute of Information Science, Academia Sinica, Taiwan 2003–2004 Graduate Research Assistant, National Taiwan University, Taiwan