Mutación o disomía uniparental? Evaluación de patrones de herencia anómalos utilizando redes bayesianas orientadas a...

Current methods of analysis in paternity testing are often based on the assumption that most incompatibilities observed between the child and the parents genetic profiles have a paternal origin. In addition, many inconsistencies are commonly treated as a product of slippage mutational events, and this premise is considered when evaluating the statistical result. Recent high-density genotype data suggest that uniparental disomy (UPD) has a higher frequency of occurrence than previously believed. Therefore, the possibility of UPD must be considered when likelihood ratios (LR) for paternity hypotheses are calculated, especially in cases where mutation events are incompatible with stepwise STR mutation models. This work proposes a statistical framework for the evaluation of paternity trio results based on Object-Oriented Bayesian Networks (OOBN) models, simultaneously considering the possibility of abnormal inheritance patterns such as mutational events and UPD. An OOBN was designed for this task using the software GeNIe Academic. A computational tool was then programmed using the python language, using pmgpy package as backend. Three million simulated paternity trios evidence, comprising all inheritance patterns modeled by the network, were used to test the model and evaluate its capacity to identify and distinguish mutation and UPD events. The proposed method can successfully assess LR values for paternity hypotheses, including statistical treatment of mutation and UPD events. For UPD occurrences, STR markers located on a single chromosome account for a unique Paternity Index (PI) value, avoiding the underestimation of paternity probability resulting from multiple mutation events. The model provides LRs to estimate the probability of occurrence of mutation or UPD based on population frequencies and case-related genetic data. All three types of UPD are included in the model, and the probability of each is provided, given the evidence. Finally, the developed computational tool can be used to process large datasets, allowing scalability of evidence inclusion and evaluation process. The proposed model provides a useful method to evaluate genetic parentage testing data when discrepancies in observed evidence might be a result of atypical inheritance situations. The statistical results are unbiased towards the paternal origin of incompatibilities and are particularly useful when results are not compatible with commonly adopted STR stepwise mutation models.