13P - A novel algorithm for predicting variant detectability in oncogenomic analysis

Background

The limit of detection (LoD) represents the minimum number of reads to detect a variant in a given depth, especially in multigene panel testing (MGPT). The reliability of MGPT depends on LoD in detecting somatic variants with precision. Comparison based LoD calculations require multiple inputs like population and non-tumorous data, which needs a high computational power. The standard MGPTs calculate in-run LoD rather than allele specific. Here, we aim to introduce a single-data-driven parallelized algorithm that overcomes these limitations.

Methods

The algorithm utilizes VCF and BAM files and derives the log-odds score from the binomial probabilities of true positive and false positive along with the variant allele fraction. Subsequently, the algorithm determines the minimum VAF threshold, indicating that the variant is confidently detectable above the specified error rate (default: 0.0001). A total of 652 liquid biopsy samples reported by a MGPT (20 genes), were used for benchmarking. We performed Fisher's exact test (p<0.05) both at the variant and gene-based level to analyze the association between true and false positive predictions.

Results

In our study, we optimized the algorithm's execution time by implementing it in a parallel processing framework. We were able to calculate LoDs in 44.32 kb/min speed within the configuration of 64 cores and 256 GB of memory. Based on reported detectable variants, the algorithm significantly demonstrated 89.92% positive predictive value (p=0.0001). When we assessed the algorithm at the gene-based level, all of 20 cancer-related genes were found statistically significant (p<0.003), which proves our testing accuracy.

Conclusions

The algorithm achieved highly significant associations by overcoming the limitations of comparison-based LoD calculations, both at the variant- and gene-based levels. These findings highlight the potential of our tool to aid precision oncogenomics from liquid biopsy samples. Overall, the study presents a novel LoD tool for all NGS applications, particularly in the scope of MGPT for somatic variants. The tool holds promise in advancing precision oncogenomics and may serve as a valuable asset in all personalized treatment strategies.

Editorial acknowledgement

Clinical trial identification

Legal entity responsible for the study

Acibadem Mehmet Ali Aydinlar University, Institute of Health Sciences, Department of Genome Studies.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.