Its not recommended to filter genes other than removing very lowly expressed genes (that could be deemed noise) or genes with very low variance (not informative). Check FAQ (Q2). If you have very few genes at hand, just make a scatterplot and calculate a spearman's correlation.
For WGCNA analysis, you could filter genes in unsupervised manner like taking top 5000 genes ranked by their variance or removing genes that do not express in more than 40% of samples etc, but not genes that belong to a pathway or genes that you are interested in (like differentially expressed). That you could do post-hoc, checking if genes of your interest are enriched in any specific WGCNA module.