應用深度神經網路進行基於短縱列重複序列的DNA混合物解譯

Abstract

短縱列重複序列 (short tandem repeat; STR) 被廣泛用於法醫學應用。EuroforMix仍然被廣泛使用來判定DNA混合物中判定其單人檢體，但在混和物中有親緣關係和裂解的混和物中，EuroforMix表現較差。 本論文的目的是探索深度神經網路 (deep neural networks) 可以幫助解決判定此類問題。為此，我們提出了基於深度神經網路的模型。如結果所示，深度神經網路在裂解及親屬混和物上可以比EuroforMix提升約29\%的準確度，整體提升約5\%的準確度。儘管裂解的檢體以及具血緣關係個體的DNA混合物在解譯上仍然具有挑戰性，但神經網路具有更高的DNA混合物解譯準確度。基於該結果，我們得出結論：有了深度神經網路，DNA混合物不必僅使用現有的軟體進行判定。

Short tandem repeat (STR) is widely used in forensic applications. EuroforMix is still widely used to determine the single DNA in the DNA mixture,However, in mixtures with related and degraded, EuroforMix performs poorly. The purpose of this paper is to explore how deep neural networks can help resolve such problems. To this end, we have proposed a model based on deep neural networks. As shown by the results, the deep neural network can provide about 29\% more accuracy than EuroforMix in degraded related mixtures, and about 5\% more accuracy overall. Although the degraded samples and DNA mixtures of related individuals are still challenging to interpret, neural networks have higher accuracy of DNA mixture interpretation. Based on this result, we conclude that with deep neural networks, DNA mixtures do not have to be judged using only existing software.