Strategies for incorporating alternative splicing variants in thyroid nodule biopsy classification based on...

Area of Science:

• Genomics
• Molecular Biology
• Cancer Diagnostics

Background:

• Accurate diagnosis of thyroid nodules with indeterminate cytology (Bethesda III and IV) is challenging.
• Current molecular tests leave some cases unresolved, leading to overtreatment or delayed care.
• Alternative splicing (AS) is prevalent and may offer new diagnostic insights.

Purpose of the Study:

• To investigate the integration of alternative splicing data with gene expression profiles for improved thyroid nodule classification.
• To evaluate the impact of different data processing and variant identification methods on classification performance.
• To validate the developed classification model using independent datasets.

Main Methods:

• Utilized gene expression data from 335 patients analyzed via HTA2.0 microarrays.
• Employed Transcriptome Analysis Console (TAC) and EventPointer for splicing variant identification.
• Performed feature selection, dataset analysis, and variant identification using a bootstrap procedure.
• Introduced modifications to deduplicate gene features.

Main Results:

• Classification performance was highly dependent on the processing methodology.
• The EventPointer pipeline showed effectiveness for gene-level features.
• Transcriptome Analysis Console (TAC)-generated variants achieved the highest bootstrap performance with 0.938 classification accuracy.
• External validation using independent datasets confirmed model performance.
• RNA-seq data verified cross-platform consistency of selected features, including thyroid cancer-associated genes like FN1 and LIPH.

Conclusions:

• Alternative splicing variants significantly influence classification quality in thyroid nodule diagnosis.
• Diagnostic classifiers can benefit from incorporating alternative splicing data for enhanced accuracy.
• This approach holds promise for improving the management of indeterminate thyroid nodules.