DeepSomatic’s ability to identify cancer-related variants was tested on breast and lung cancer genomes from the CASTLE dataset. Training included three breast cancer genomes and two lung cancer genomes. Performance was evaluated on an unseen breast cancer genome and chromosome 1 from all samples. Results indicated that DeepSomatic outperformed other methods, effectively detecting tumor variants with high accuracy. It surpassed tools like SomaticSniper, MuTect2, and Strelka2 in short-read sequencing, while outperforming ClairS in long-read sequencing. DeepSomatic identified 329,011 somatic variants across six cell lines and a preserved sample, excelling in identifying insertions and deletions (Indels). The F1-score increased significantly, with DeepSomatic achieving 90% on Illumina sequencing for Indels, compared to 80% for the next best method. On Pacific Biosciences data, DeepSomatic scored over 80%, while competitors fell below 50%. These results affirm DeepSomatic’s superior performance in cancer genomics.
Source link
Share
Read more