Large scale genetic identity inference using probabilistic model checking

The identification of accident victims uses techniques such as visual recognition, fingerprint or dental records comparison. In case of disasters, however, where the number of victims is large, the use of these techniques becomes unfeasible, because they require the comparison of ante mortem with post mortem information, and it is often not possible to obtain the former one. An alternative to these traditional methods is DNA identification. It is an accurate method that requires the acquisition of DNA samples from victims and their relatives. After samples are sequenced, a Bayesian network can be constructed to determine the likelihood that the victim is indeed part of that family. While accurate, this process is computer intensive, and quickly becomes too expensive to perform as the number of victims grows. In the case of large accidents such as the 2004 Tsunami or the recent tragedies in Haiti, a complete identification of the victims using DNA comparison would take decades to perform. In this paper we propose the use of probabilistic model checking techniques to alleviate this problem. The model compares allele values between different individuals, verifying if these values are in accordance with Mendelian genetic law. If a Mendelian rule is violated, the victim cannot belong to that family. With our approach we are able to calculate the probability of someone belong to a family taking less than one minute. For comparison, determining the relationship between a single victim and its family using a Bayesian network takes about 5 minutes using an automated expert system. For 1,000 victims and families, about half a million networks would have to be computed, adding up to about 5 years of CPU time to reach the same result!