Issue 01AI 바이오 논문
AC POST
AI 바이오 논문 목록
arxiv2026년 9월 20일 09:00

miRstring: 종을 초월한 성숙 miRNA 해독 및 인공 소형 RNA 설계를 가능하게 하는 RNA 언어 모델

MicroRNAs (miRNAs)는 구조화된 전구체로부터 가공된 후 Argonaute 단백질에 결합하여 표적 mRNAs를 억제한다. 그러나 전구체 맥락에서 성숙한 miRNAs를 해독할 수 있는 일반화된 계산 프레임워크는 여전히 제한적이며, 이는 종 간 주석 달기(cross-species annotation)와 합리적인 인공 miRNA 설계를 모두 제약하고 있다. 본 연구에서는 miRNA/miRNA* 이중 가닥을 정의하는 네 가지 경계를 해독하는 생성 인식 RNA 언어 프레임워크인 miRstring을 제시한다. 414개 종에 걸친 77,708개의 miRNA 전구체로 학습된 miRstring은 family- 및 species-held-out 평가에서 기존 방법보다 뛰어난 성능을 보였으며, 성숙한 miRNAs의 첫 번째 뉴클레오타이드를 정확하게 식별한다. 중요한 점은, 이 모델의 attention mechanism이 endonuclease에 의해 절단되는 miRNA/miRNA* 경계 부위를 강조한다는 것이며, 이는 모델이 생물학적으로 의미 있는 특징을 포착하고 있음을 나타낸다. 나아가, miRstring을 사용하여 인공 miRNAs를 위한 최적의 pre-miRNA 스캐폴드를 설계하였으며, 표적 mRNAs를 억제하는 데 있어 그 효능을 검증하였다. 종합적으로 miRstring은 종 간 성숙-miRNA 주석 달기부터 예측 설계에 이르는 확장 가능한 경로를 구축하여, 인공지능 기반의 miRNA 엔지니어링을 가능하게 하고 계산 능력을 확장한다.

MicroRNAs (miRNAs) are processed from structured precursors and subsequently loaded into Argonaute proteins to repress target mRNAs. Yet generalized computational frameworks capable of decoding mature miRNAs from precursor context remain limited, constraining both cross-species annotation and rational artificial miRNA design. Here, we present miRstring, a biogenesis-aware RNA language framework that decodes the four boundaries defining the miRNA/miRNA duplex. Trained on 77,708 miRNA precursors spanning 414 species, miRstring outperforms existing methods under family- and species-held-out evaluations and accurately identifies the first nucleotide of mature miRNAs. Importantly, its attention mechanism highlights the miRNA/ miRNA* boundary sites cleaved by endonucleases, indicating that the model captures biologically meaningful features. Furthermore, we employed miRstring to design optimal pre-miRNA scaffolds for artificial miRNAs and validated its efficacy in repressing target mRNAs. Taken together, miRstring establishes a scalable route from cross-species mature-miRNA annotation to predictive design, enabling artificial intelligence-driven miRNA engineering and extending computatio