Development And Application Of A Hidden Markov Model Based Method For Somatic CNVs Detection At Single Cell Level

Copy Number Variation(CNVs),which is defined as the events of DNA fragment usually larger than 50bp with variable copy numbers compared to reference genome,is one of the most common variation types.The emergence of CNVs is the adaptive consequence during normal biology,and is also frequently occurred in tumors.Given that next generation sequencing technology has improved the detection throughput and power of CNVs compared to array methods,an increasing accumulation of research has evidenced somatic CNVs play an important role in driving the initiation and development of tumor.The development of single cell sequencing technology provides potential to explore the diversity of CNVs within one tumor.However,the limited DNA amount and bias introduced primarily by whole genome amplification process challenge the CNVs detection used in tumor bulk,and CNVs detection methods for single cells has not been well established.Here,we first developed a somatic CNV detection method based on Hidden Markov Model(HMM)algorithm,including correction for GC and mappability bias,identification of CNVs according to sequencing reads distribution and segmentation of adjacent windows harboring comparable copy number states.Next,we simulated-0.3X next generation sequencing data of chromosome 12,and then spiked into CNVs with different length,from 10k to 2M.We used detection windows with different length(10k,30k,50k,100k)and performed somatic CNVs detection using our model.We found that our method had higher detection efficiency of CNVs larger than 500k,especially those far larger than the detection window,and could achieve remarkable detection sensitivity higher than 85%and specificity around 95%for large CNVs(>500k).Finally,we applied this method to detect somatic CNVs of 50 single cells(around 0.3X for each)derived from same glioblastoma multiforme.The remarkable correlation of somatic CNVs pattern between single cells and tumor bulk suggested the robustness of our method.Besides,common somatic CNVs observed in almost tumor single cells like amplification on chr7(harboring EGFR)demonstrated early event that may drive tumor initiation,while recurrent events occurred in partial tumor single cells like deletion on chr9(harboring CDKN2A/B)suggested later event that may drive the tumor evolution.Heterogeneous somatic CNVs pattern provides evidence to trace the dynamic evolution and will guide further targeted therapy.Even though our method still has limitation when detecting focal CNVs or for ultra-low sequencing data,our work will not only give insights into developing robuster tools for CNVs detection at low-sequenced single cell level,but also provide methods to decipher heterogeneous somatic CNVs pattern,making us a better understanding of tumor dynamic evolution and facilitating the tumor precision medicine research.