Kinship analysis and machine learning algorithms in forensic contexts: A new NGS panel

Establishing kinship relationships plays a crucial role in solving inheritance disputes, criminal investigations, identifying disasters or war victims, and searching for missing person. Short Tandem Repeats (STRs) are the gold-standard markers and can accurately identify paternity and first-degree relationships. However, they have limited effectiveness in identifying second-degree and more distant kinships due to high mutation rates and the limited number of markers contained in commercial kits. Therefore, a novel panel of 4849 Single Nucleotide Polymorphisms (SNPs) was created for high-resolution kinship testing. Over 150,000 pairs of simulated individuals, covering unrelated and related cases up to the 5th degree of relatedness, were analyzed. A combination of the Forrel package (R) and a supervised machine learning approach was evaluated for data interpretation. Our panel demonstrated remarkable abilities to accurately identify kinship relationships, even at higher degrees, such as 3rd and 4th degree. Despite observing low precision values for 5th-degree relationships inference, promising performance with recall values surpassing 0.6 was achieved. Integrating supervised machine learning algorithms further improved the panel's performance, especially in the inference of complex kinship relationships. In particular, the F1 score showed an improvement of approximately 12.25 % and 20 % for 4th-degree and 5th-degree relationships, respectively. Additionally, a significant strength of this method is its accuracy of more than 99 % in effectively distinguishing unrelated from related pairs. Therefore, this new panel could pave the way for a significantly improved and comprehensive method of kinship relationship inference.