MR analyses in which the genetic IV-risk factor association and genetic IV-outcome association are from the same sample and individual-level data is used to derive the MR estimate.
Genetic IVs must fulfil the same IV assumptions. The gene-outcome model must be correctly specified, but the gene-exposure model can be miss-specified. One-sample MR is more prone to data overfitting than two-sample MR. Though, two-sample MR may be similar to one-sample if there is overlap of participants in the two samples. In one-sample MR, weak instrument bias will tend to bias towards the confounded multivariable regression estimate.