Single nucleotide polymorphisms (SNPs) are the most common form of genetic variation amongst the human population and are key to personalized medicine. New tests are presented to distinguish pathogenic/malign (i.e., likely to contribute to or cause a disease) from nonpathogenic/benign SNPs, regardless of whether they occur in coding (exon) or noncoding (intron) regions in the human genome. The tests are based on the nearest neighbor (NN) model of Gibbs free energy landscapes of DNA hybridization and on deep structural properties of DNA revealed by an approximating metric (the h-distance) in DNA spaces of oligonucleotides of a common size. The quality assessments show that the newly defined PNPG test can classify a SNP with an accuracy about 73% for the required parameters. The best performance among machine learning models is a feed-forward neural network with fivefold cross-validation accuracy of at least 73%. These results may provide valuable tools to solve the SNP classification problem, where tools are lacking, to assess the likelihood of disease causing in unclassified SNPs. These tests highlight the significance of hybridization chemistry in SNPs. They can be applied to further the effectiveness of research in the areas of genomics and metabolomics.

单核苷酸多态性 (SNP) 是人群中最常见的遗传变异形式，是个性化医疗的关键。提出了新的测试来区分致病性/恶性（即可能导致或引起疾病）与非致病性/良性 SNP，无论它们是出现在人类基因组的编码（外显子）还是非编码（内含子）区域。这些测试基于 DNA 杂交吉布斯自由能图的最近邻 (NN) 模型和近似度量（h-距离）在相同大小的寡核苷酸的 DNA 空间中。质量评估表明，新定义的 PNPG 测试可以对所需参数的 SNP 进行分类，准确率约为 73%。机器学习模型中最好的性能是前馈神经网络，其五倍交叉验证准确度至少为 73%。这些结果可能为解决 SNP 分类问题提供有价值的工具，在缺乏工具的情况下，评估未分类 SNP 引起疾病的可能性。这些测试突出了 SNP 中杂交化学的重要性。它们可用于进一步提高基因组学和代谢组学领域研究的有效性。