Analytical Validation
Erroneous results
The definition of acceptable error rates is based on the inverse of the established minimum performance thresholds for well defined standards. The procedure is detailed in the Analytical Performance Evaluation Plan (Doc ID: FRK-PEP-02).
The following parameters, each of which should be evaluated independently, define erroneous results:
Sensitivity (Indel): < 0.992
Sensitivity (SNP): <0.999
Sensitivity (CNV deletion):<0.99 for deletion, <0.98 for duplication (variant and prediction levels)
Precision (Indel): < 0.992
Precision rate (SNP): <0.997
Precision rate (CNV):< 0.99 for deletions and duplications (variant and prediction levels)
Overall F-score (SNP and Indel): <99.5%
Overall F-score (CNV): <99% for deletions, < 98.4% for duplications (variant and prediction levels)
Threshold for significant increase in errors:
Threshold for deviation from the baseline error rate that should be considered significant are presented below:
Parameter | Threshold (%) |
Sensitivity (Indel) | 0.25 |
Sensitivity (SNP) | 0.06 |
Sensitivity (CNV) | 0.50 |
Precision (Indel) | 0.25 |
Precision rate (SNP) | 0.01 |
Precision rate (CNV) | 0.50 |
Clinical Validation
Erroneous results
Expected erroneous results in clinical validation are defined by the system’s ability to correctly annotate, filter, and present variants in concordance with established standards for variant annotation and interpretation. The acceptance criteria are established in the Clinical Performance Evaluation Plan (Doc ID: FRK-PEP-03). Erroneous results are defined as performance falling below the predefined acceptance thresholds for the following core software parameters:
Annotation accuracy
Gene concordance: < 99%
Region classification accuracy: < 95%
Effect classification accuracy: < 95%
HGVS nomenclature concordance (c. and p. notation): < 95%
Cross-build consistency (GRCh37 vs GRCh38): < 99%
Filtering performance
Filter sensitivity (per filter): < 99%
Filter specificity (per filter): < 99%
Family segregation
Inheritance pattern concordance: < 100%
Threshold for significant increase in errors
Threshold for deviation from the baseline error rate that should be considered significant are presented below:
Parameter | Threshold (%) |
Gene concordance | 0.5 |
Region classification | 2.0 |
Effect classification | 2.0 |
HGVS concordance (c./p.) | 2.0 |
Cross-build consistency | 1.0 |
Filter sensitivity | 0.5 |
Filter specificity | 0.5 |
Segregation concordance | 0.5 |