Good question! A popular motif (and perhaps the most illustrative example) is to restrict the instrument to be genotype variants within a specific gene, and us

Good question! A popular motif (and perhaps the most illustrative example) is to restrict the instrument to be genotype variants within a specific gene, and use its protein product (for example, its serum concentration) as the exposure [1]. In this way, horizontal pleiotropy is significantly mitigated. This is sort of a natural extension (on a statistical methodological level, really a simplification) of a popular technique where a drug which has a known protein target is used in a "factorial MR" scheme [2]. Unfortunately, I'm less familiar with the literature here because it's very domain-specific. [1] doi:10.2337/dc18-2444 [2] doi:10.1016/j.jacc.2015.02.020 Edit: removed one of the references because it's a somewhat convoluted example, and on a deeper read, wasn't actually the paper I was thinking of.