A benchmarking framework for the accurate and cost-effective detection of clinically-relevant structural variants for cancer target identification and diagnosis

期刊论文
出版模式：

开放获取
发表日期：

2024年1月16日

查看全文

Guiwu Zhuang,
Xiaotao Zhang,
Wenjing Du,
Libin Xu,
Jiyong Ma,
Haitao Luo,
Hongzhen Tang,
Wei Wang,
Peng Wang,
Miao Li,
Xu Yang,
Dongfang Wu,
Shencun Fang

摘要

Background

Accurate clinical structural variant (SV) calling is essential for cancer target identification and diagnosis but has been historically challenging due to the lack of ground truth for clinical specimens. Meanwhile, reduced clinical-testing cost is the key to the widespread clinical utility.

Methods

We analyzed massive data from tumor samples of 476 patients and developed a computational framework for accurate and cost-effective detection of clinically-relevant SVs. In addition, standard materials and classical experiments including immunohistochemistry and/or fluorescence in situ hybridization were used to validate the developed computational framework.

Results

We systematically evaluated the common algorithms for SV detection and established an expert-reviewed SV call set of 1,303 tumor-specific SVs with high-evidence levels. Moreover, we developed a random-forest-based decision model to improve the true positive of SVs. To independently validate the tailored ‘two-step’ strategy, we utilized standard materials and classical experiments. The accuracy of the model was over 90% (92–99.78%) for all types of data.

Conclusion

Our study provides a valuable resource and an actionable guide to improve cancer-specific SV detection accuracy and clinical applicability.