Abstract
BACKGROUND: Tumor Mutation Burden (TMB) is commonly characterized as the number of non-synonymous somatic SNVs per megabase within the gene region identified through whole exon sequencing or targeted sequencing in a tumor sample. It has been statistically demonstrated that TMB was related to the ability of neoantigen production and used to predict the efficacy of immunotherapy for various types of cancers. However, screening for TMB in patients poses challenges due to the extensive labor and financial resources required for the preparation of large quantities of parallel sequencing libraries. RESULTS: In this study, we employed compressed sensing (CS) to calculate TMB from overlapped pooling sequencing data, aiming to reduce the sequencing cost by minimizing the number of library builds. Over 90% SNPs could still be detected without a significant loss of mutation information even when the data is pooled from ten different samples. Based on this, the orthogonal matching pursuit (OMP) algorithm and the basic pursuit (BP) algorithm were used to reconstruct TMB from pooling sequencing data. The performance of these two algorithms was evaluated. The BP algorithm consistently performed well across all cases, albeit necessitating extended computational time. The OMP algorithm has been proved to be suitable for scenarios where the original matrix was sparse but it showed low overall performance. Based on an accurate calculation of TMB, we determined that the number of sequencing runs could be reduced to 0.6 times the total number of samples, resulting in a 40% reduction in sequencing cost. CONCLUSIONS: In conclusion, we calculated TMB from overlapped pooling sequencing data utilizing compressed sensing strategy to reduce sequencing cost. Our findings confirm that the SNP calling from ten samples' pooling sequencing data is feasible. Additionally, we performed an assessment of the reconstruction efficiency of both the BP model and the OMP model.