It is very easy to find correlations between phenotypic traits (getting MS, developing cancer, autism etc) and "your favourite biological variable". See John Ioannidis talking about microRNA studies for example. Part of the reason for this is that high throughput data like gene expression measures, gene methylation or microbiome sequencing gives you p << n data which invariably is also low dimensional ie a few factors explain most of thr variation. It is therefore easy to find a variable or a 'signature' which correlates with one of these explanatory factors and the phenotype, but doesn't tell you anything much about what is going on, or how to treat the disease. But it does allow thousands of papers to be published with pvalues <0.05.

Not to downplay this data, but caution is required not to over interpret the results, and avoid making the sane mistakes over and over again.