In this work, Minikel et al. exploit predicted loss-of-function (pLoF) data from the Genome Aggregation Database (gnomAD), including information from 141,456 individuals, to assess how genetic variation can inform drug target selection. Essential genes are likely to accumulate less loss of function variants (i.e. be more constrained) in the human population compared to non-essential genes. The authors first assess to what degree current drug targets are affected by this constraint. They infer this by computing a measure of essentiality that is a ratio between the pLoF variants found in a gene and the expected number of variants based on DNA mutation rates. Although they observe that current drug targets are significantly more constrained (i.e. have lower observed vs. expected ratios) than the overall protein-coding genes, they also see a wide degree of variability in constraint values among drug targets. If they consider drug indications, they observe that drugs used in cancer treatment show a high degree of constraint compared to other drugs. This would be expected, as genes coding for such drug targets are usually essential genes targeted by cytotoxic drugs. This interpretation is more complicated for other indications as in the case of highly constrained targets for drugs treating neurological diseases. In this case, according to the authors, these targets exemplify how a gene may essentiality be restricted to particular time points in the lifespan of an individual (e.g. during embryological development), and inhibition may still be well tolerated in adults. Overall, ~20% of the analyzed drug targets have lower observed vs. expected values than the average found for genes known to cause severe diseases of haploinsufficiency. Therefore, these results speak against considering constraint as a hard filter for target druggability. Furthermore, the authors also assess the likelihood of detecting ‘human knockouts’ for particular genes of interest in their dataset as these knockouts could reveal the lifelong biological impact of the inhibition of potential drug targets. According to their calculations, although pLoF heterozygosity already exists in around 80% of genes in gnomAD, detecting total knockout individuals in outbred populations would require a 1000-fold larger sample size and, in some cases, would never be attained. However, sequencing currently under-sampled populations, including more data on bottlenecked populations, and especially focusing on consanguineous individuals could enhance the possibilities of detecting tolerated complete knockouts of interest. In summary, this analysis calls for a careful assessment of gene variation in candidate drug targets. It also provides useful estimates on the work that still needs to be done to ensure data on genomic variability can be used to obtain a deeper understanding of gene function and to select new effective drug targets and drug indications.