Estimation of tumor purity is a crucial step in understanding the complex copy number landscapes of heterogeneous tumor samples through next-generation sequencing data. A prominent problem in tumor purity estimation is how to disambiguate multiple compatible solutions, as distinct combinations of tumor purity and ploidy can fit the sequencing data equivalently well. Current methods "break the tie" relying on either prior information or numerical heuristics rather than comprehensively explaining the sequencing data itself. Such approaches will lead to fundamental bias when the ground truth significantly deviates from the prior information. Here we propose a full probabilistic model-based method PyLOH that leverage the cluster pattern of loss of heterozygosity observed in paired cancer sequencing data to disambiguate multiple compatible solutions. We also introduce a novel visualization method "BAF heat map" to to characterize the cluster pattern of LOH.