Department of Microbiology, University of Tennessee-Knoxville, Knoxville, TN, United States of America.
Department of Anthropology, University of Tennessee-Knoxville, Knoxville, TN, United States of America.
PLoS One. 2024 Oct 11;19(10):e0311906. doi: 10.1371/journal.pone.0311906. eCollection 2024.
Microbial succession has been suggested to supplement established postmortem interval (PMI) estimation methods for human remains. Due to limitations of entomological and morphological PMI methods, microbes are an intriguing target for forensic applications as they are present at all stages of decomposition. Previous machine learning models from soil necrobiome data have produced PMI error rates from two and a half to six days; however, these models are built solely on amplicon sequencing of biomarkers (e.g., 16S, 18S rRNA genes) and do not consider environmental factors that influence the presence and abundance of microbial decomposers. This study builds upon current research by evaluating the inclusion of environmental data on microbial-based PMI estimates from decomposition soil samples. Random forest regression models were built to predict PMI using relative taxon abundances obtained from different biological markers (bacterial 16S, fungal ITS, 16S-ITS combined) and taxonomic levels (phylum, class, order, OTU), both with and without environmental predictors (ambient temperature, soil pH, soil conductivity, and enzyme activities) from 19 deceased human individuals that decomposed on the soil surface (Tennessee, USA). Model performance was evaluated by calculating the mean absolute error (MAE). MAE ranged from 804 to 997 accumulated degree hours (ADH) across all models. 16S models outperformed ITS models (p = 0.006), while combining 16S and ITS did not improve upon 16S models alone (p = 0.47). Inclusion of environmental data in PMI prediction models had varied effects on MAE depending on the biological marker and taxonomic level conserved. Specifically, inclusion of the measured environmental features reduced MAE for all ITS models, but improved 16S models at higher taxonomic levels (phylum and class). Overall, we demonstrated some level of predictability in soil microbial succession during human decomposition, however error rates were high when considering a moderate population of donors.
微生物演替被认为可以补充现有的人死后时间（PMI）估计方法。由于昆虫学和形态学 PMI 方法的局限性，微生物作为法医应用的一个有趣目标，因为它们存在于所有分解阶段。以前从土壤坏死生物群落数据中建立的机器学习模型产生的 PMI 误差率为两天半到六天；然而，这些模型仅基于生物标志物（例如 16S、18S rRNA 基因）的扩增子测序构建，并且不考虑影响微生物分解者存在和丰度的环境因素。本研究通过评估包含环境数据对来自分解土壤样本的基于微生物的 PMI 估计的影响，在此基础上进行了研究。随机森林回归模型被构建用于使用从不同生物标志物（细菌 16S、真菌 ITS、16S-ITS 组合）和分类水平（门、纲、目、OTU）获得的相对分类群丰度来预测 PMI，同时包括和不包括来自 19 名在土壤表面分解的已故人类个体的环境预测因子（环境温度、土壤 pH 值、土壤电导率和酶活性）（田纳西州，美国）。通过计算平均绝对误差（MAE）来评估模型性能。所有模型的 MAE 范围为 804 到 997 个累积度日（ADH）。16S 模型的性能优于 ITS 模型（p = 0.006），而将 16S 和 ITS 结合起来并没有提高 16S 模型的性能（p = 0.47）。在 PMI 预测模型中包含环境数据对 MAE 的影响因保守的生物标志物和分类水平而异。具体来说，包含测量的环境特征降低了所有 ITS 模型的 MAE，但提高了较高分类水平（门和纲）的 16S 模型的 MAE。总体而言，我们在一定程度上证明了人类分解过程中土壤微生物演替的可预测性，但是当考虑到中等数量的供体时，误差率很高。
