Abstract
BACKGROUND/OBJECTIVES: Kinship inference is commonly adopted in various forensic applications, but previous studies have often lacked precision. METHODS: In this study, a new method for the nomenclature of kinship types, i.e., kinship chain (KC), was proposed, and then, six types of identity by state (IBS) scores were calculated for simulated and real families using four types of markers. Finally, several Bayesian network (BN)-based classifiers were constructed to investigate the efficiency of the kinship inference. RESULTS: A total of 7, 22, 58, and 3 KCs were obtained for common first-, second-, and third-degree relatives and unrelated pairs, respectively. High accuracies could be achieved in distinguishing between related and unrelated pairs after combining the four types of genetic markers, with an accuracy of >99.99% for all 7 KCs of first-degree relationships and ~99% for 14 out of 22 KCs of second-degree relatives. When comparing relationships of the same degree, the accuracies were 99.28%, 42.31%, and 15.82% for first-, second-, and third-degree relationships, respectively. When it came to differentiating unspecific relationships, the overall accuracy was over 80%. All the results were validated on real family data. CONCLUSIONS: With the new nomenclature method of kinship types and the combination of autosomal and non-autosomal genetic markers, kinship inference can be realized with high accuracy and precision, which will be helpful in complex forensic cases, such as the identification of mass disaster victims.