Benchmarking Fine-Tuned RNA Language Models for Branch Point Prediction

Accurately predicting branch points in RNA splicing is fundamental to understanding splicing mechanisms and identifying pathogenic genetic variants. In this work, we fine-tuned and evaluated several RNA language models for branch point detection, achieving state-of-the-art performance. The top-performing model, ERNIE-RNA, achieved an $F_1$ score of 0.811, sequence accuracy of 0.790, and an AP score of 0.868, surpassing the performance of previous leading models. Our findings suggest that hyperparameter optimization and extended training could significantly increase performance, establishing this work as a baseline for future research.