Bastida-Lertxundi Nerea, Martí-Carrera Itxaso, Laña-Ruíz Borja, Martínez-Múgica Barbosa Otilia, Muguerza-Iraola Raquel, Sáez-Villaverde Raquel, Crettaz Julien S
Biogipuzkoa Health Research Institute, Neurogenetics, Biology and RNA Therapies Research Group - NeuroRNA, San Sebastián, Spain.
Osakidetza, Donostialdea Integrated Health Organization, Unit of Clinical Genetics, Donostia University Hospital, San Sebastián, Spain.
Adv Lab Med. 2025 Mar 10;6(1):28-36. doi: 10.1515/almed-2025-0031. eCollection 2025 Mar.
To evaluate seven bioinformatics platforms for automated AI-based genomic variant prioritization and classification.
An evaluation was performed of 24 genetic variants that explained the phenotype of 20 patients. FASTQ files were simultaneously uploaded on the following bioinformatics platforms: Emedgene, eVai, Varsome Clinical, CentoCloud, QIAGEN Clinical Insight (QCI) Interpret, SeqOne and Franklin. Automated variant prioritization and classification was performed using patient phenotypes. Phenotypes were entered onto the different platforms using HPO terms. The classification of reference was established based on the criteria of the American College of Medical Genetics and Genomics (ACMG) and the Association of Molecular Pathology and ACMG/ClinGen guidelines.
SeqOne demonstrated the highest performance in variant prioritization and ranked 19 of 24 variants in the Top 1; four in the Top 5, and one in the Top 15, followed by CentoCloud and Franklin. QCI Interpret did not prioritize six variants and failed to detect one. Emedgene did not prioritize one and failed to detect one. Finally, Varsome Clinical did not prioritize four variants. Franklin classified correctly 75 % of variants, followed by Varsome Clinical (67 %) and QCI Interpret (63 %).
SeqOne, CentoCloud, and Franklin had the highest performance in automated variant prioritization, as they prioritized all variants. In relation to automated classification, Franklin showed a higher concordance with the reference and a lower number of discordances with clinical implications. In conclusion, Franklin emerges as the platform with the best overall performance. Anyway, further studies are needed to confirm these results.
评估七个基于人工智能的生物信息学平台用于自动进行基因组变异优先级排序和分类的情况。
对24个解释20名患者表型的基因变异进行评估。FASTQ文件同时上传至以下生物信息学平台：Emedgene、eVai、Varsome Clinical、CentoCloud、QIAGEN Clinical Insight（QCI）Interpret、SeqOne和Franklin。使用患者表型进行自动变异优先级排序和分类。使用人类表型本体（HPO）术语在不同平台上输入表型。参考分类依据美国医学遗传学与基因组学学会（ACMG）以及分子病理学协会和ACMG/临床基因组资源（ClinGen）指南的标准来确定。
SeqOne在变异优先级排序方面表现最佳，24个变异中有19个排在前1；4个排在前5，1个排在前15，其次是CentoCloud和Franklin。QCI Interpret未对6个变异进行优先级排序且未能检测到1个变异。Emedgene未对1个变异进行优先级排序且未能检测到1个变异。最后，Varsome Clinical未对4个变异进行优先级排序。Franklin正确分类了75%的变异，其次是Varsome Clinical（67%）和QCI Interpret（63%）。
SeqOne、CentoCloud和Franklin在自动变异优先级排序方面表现最佳，因为它们对所有变异都进行了优先级排序。在自动分类方面，Franklin与参考的一致性更高，与临床意义的不一致性数量更少。总之，Franklin是整体性能最佳的平台。无论如何，需要进一步研究来证实这些结果。
