Large-scale analysis reveals splicing biomarkers for tuberculosis progression and prognosis

• Published in: Computers in biology and medicine Vol. 171; p. 108187
• Main Authors: Lai, Hongli, Lyu, Mengyuan, Ruan, Hongxia, Liu, Yang, Liu, Tangyuheng, Lei, Shuting, Xiao, Yuling, Zhang, Shu, Ying, Binwu
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01-03-2024; Elsevier Limited
• Subjects: Alternative splicing; Biomarkers; Datasets; Disease; Exons; Gene expression; Machine learning; Medical prognosis; Performance evaluation; Performance prediction; Prediction models; Prognosis; Progression; Tuberculosis; Progression; Alternative splicing; Tuberculosis; Prognosis
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2024.108187

Emerging evidence suggests that aberrant alternative splicing (AS) may play an important role in tuberculosis (TB). However, current knowledge regarding the value of AS in TB progression and prognosis remains unclear.
Public RNA-seq datasets related to TB progression and prognosis were searched and AS analyses were conducted based on SUPPA2. Percent spliced in (PSI) was used for quantifying AS events and multiple machine learning (ML) methods were employed to construct predictive models. Area under curve (AUC), sensitivity and specificity were calculated to evaluate the model performance.
A total of 1587 samples from 7 datasets were included. Among 923 TB-progression related differential AS events (DASEs), 3 events (GET1-skipping exon (SE), TPD52-alternative first exons (AF) and TIMM10-alternative 5′ splice site (A5)) were selected as candidate biomarkers; however, their predictive performance was limited. For TB prognosis, 5 events (PHF23-AF, KIF1B-SE, MACROD2-alternative 3’ splice site (A3), CD55-retained intron (RI) and GALNT11-AF) were selected as candidates from the 1282 DASEs. Six ML methods were used to integrate these 5 events and XGBoost outperformed than others. AUC, sensitivity and specificity of XGBoost model were 0.875, 81.1% and 83.5% in training set, while they were 0.805, 68.4% and 73.2% in test set.
GET1-SE, TPD52-AF and TIMM10-A5 showed limited role in predicting TB progression, while PHF23-AF, KIF1B-SE, MACROD2-A3, CD55-RI and GALNT11-AF could well predict TB prognosis and work as candidate biomarkers. This work preliminarily explored the value of AS in predicting TB progression and prognosis and offered potential targets for further research.
•Three AS events (GET1-SE, etc.) were selected as candidate markers of TB progression.•For TB prognosis, 5 events (PHF23-AF, KIF1B-SE, etc.) were identified as candidates.•XGBoost model integrating AS events performed well in predicting TB prognosis.