Search

Summary: The serine-/arginine-rich splicing factor 2 (SRSF2) plays pivotal roles in pre-mRNA processing and gene transcription. Recurrent mutations, particularly a proline-to-histidine substitution at position 95 (P95H), are common in neoplastic diseases. Here, we assess SRSF2’s diverse functions in squamous cell carcinoma. We show that SRSF2 deletion or homozygous P95H mutation both cause extensive DNA damage leading to cell-cycle arrest. Mechanistically, SRSF2 regulates efficient bi-directional transcription of DNA replication and repair genes, independent from its function in splicing. Further, SRSF2 haploinsufficiency induces DNA damage without halting the cell cycle. Exposing mouse skin to tumor-promoting carcinogens enhances the clonal expansion of heterozygous Srsf2 P95H epidermal cells but unexpectedly inhibits tumor formation. To survive carcinogen treatment, Srsf2 P95H+/− cells undergo substantial transcriptional rewiring and restore bi-directional gene expression. Thus, our study underscores SRSF2’s importance in regulating transcription to orchestrate the cell cycle and the DNA damage response.

Abstract Direct RNA sequencing offers the possibility to simultaneously identify canonical bases and epi-transcriptomic modifications in each single RNA molecule. Thus far, the development of computational methods has been hampered by the lack of biologically realistic training data that carries modification labels at molecular resolution. Here, we report on the synthesis of such samples and the development of a bespoke algorithm, mAFiA (m6A Finding Algorithm), that accurately detects single m6A nucleotides in both synthetic RNAs and natural mRNA on single read level. Our approach uncovers distinct modification patterns in single molecules that would appear identical at the ensemble level. Compared to existing methods, mAFiA also demonstrates improved accuracy in measuring site-level m6A stoichiometry in biological samples.

Summary: The human transcriptome predominantly consists of noncoding RNAs (ncRNAs), transcripts that do not encode proteins. The noncoding transcriptome governs a multitude of pathophysiological processes, offering a rich source of next-generation biomarkers. Toward achieving a holistic view of disease, the integration of these transcripts with clinical records and additional data from omic technologies (“multiomic” strategies) has motivated the adoption of artificial intelligence (AI) approaches. Given their intricate biological complexity, machine learning (ML) techniques are becoming a key component of ncRNA-based research. This article presents an overview of the potential and challenges associated with employing AI/ML-driven approaches to identify clinically relevant ncRNA biomarkers and to decipher ncRNA-associated pathogenetic mechanisms. Methodological and conceptual constraints are discussed, along with an exploration of ethical considerations inherent to AI applications for healthcare and research. The ultimate goal is to provide a comprehensive examination of the multifaceted landscape of this innovative field and its clinical implications.