ProtSSSD, our work on protein disorder region prediction, has been published in IEEE International Conference on Bioinformatics and Biomedicine 2025.

Determining the intrinsically disordered regions of proteins is crucial for elucidating their biological functions and the mechanisms of related diseases. However, current single-sequence-based methods have low accuracy, while evolution-profile-based methods are computationally intensive. Therefore, we propose a highly accurate protein disorder predictor, ProtSSSD, which utilizes the protein language model along with a LoRA fine-tuning strategy and incorporates a multitask learning architecture. The results show that ProtSSSD outperforms all single-sequence-based methods and achieves better performance than two other language model-based techniques in the second Critical Assessment of protein Intrinsic Disorder prediction experiment. Moreover, ProtSSSD demonstrates comparable or even superior performance to the state-of-the-art profile-based method SPOT-Disorder2.

中文简介

确定蛋白质的固有无序区域对于阐明其生物学功能和相关疾病的机制至关重要。然而，目前基于单序列的方法准确率较低，而基于进化谱的方法计算量巨大。因此，我们提出了一种高精度的蛋白质无序预测器ProtSSSD，它利用蛋白质语言模型，结合LoRA微调策略，并融入了多任务学习架构。结果表明，ProtSSSD的性能优于所有基于单序列的方法，并且在第二次蛋白质固有无序预测关键评估实验中，其性能优于其他两种基于语言模型的技术。此外，ProtSSSD的性能与目前最先进的基于进化谱的方法SPOT-Disorder2相当甚至更优。