Inferring direction of therapeutic effect implied by coding genetic evidence

How would you all suggest Open Targets be used to infer whether or not a statistical association between a gene and a disease suggests an opportunity for an agonistic or antagonistic therapeutic?

It seems to me that between current in silico predictions of things like variant pathogenicity [e.g. AlphaMisssense], expression [e.g. Enformer], and regulatory function [3|Flatiron Institute - Sei] it ought to at least be possible to attempt to predict whether coding variants (particularly missense) confer a loss or gain of function. The method in [3] explicitly does this for noncoding variants, so presumably it’s even easier for coding variation.

If Getting signs of GWAS effects relative to major alleles instead of alt alleles was also possible via Open Targets, then it doesn’t seem implausible that you could use the two of these to say something about the likely mechanism of action implied by a given coding variant, in a given population.

The paper Systematic disease-agnostic identification of therapeutically actionable targets using the genetics of human plasma proteins from you all was exciting in that this kind of implied directionality is implicit in 2 sample MR. The extent to which something similar is possible now with previous GWAS evidence is not clear to me. Any thoughts would be much appreciated!

Also, I would have liked to have included more links/references in that post but kept hitting:

Screen Shot 2023-09-28 at 11.41.27 AM

This is pretty maddening for new people IMO and I know it can be disabled via the newuser max links property under AdminSettingsPosting. I’d recommend increasing it.

Hi @eczech, sorry you had to face this issue. It was probably a default setting and has been modified. You should be able to add more links to your post now. Thanks.

In terms of the analysis provided by Open targets genetics, the colocalisation results can provide some insight in directionality, for example, at the following liver enzyme locus , the beta is negative, there is a strong colocalisation with the ANKA gene, which has a negative QTL beta. indicating that lower levels of ANKA is associated with lower liver enzyme levels. However, This cannot infer causality, where additional analysis, like MR would be required.

Best wishes,

Thanks Xiangyu, that makes sense but it’s not quite what I’m after.

A part of this ask that is less typical is that I’m solely concerned with coding associations. I would be fine taking putative LoF variants or in silico functionalizations, e.g. from Sei, of missense variants (as GoF or LoF) as indications of how these variants effect the gene they code for.

This is a vastly smaller pool of variants I’m talking about, but they’re far more predictive of clinical outcomes, much easier to infer causality for and there are quite a lot of them now so I’m seeing what I think is a good reason in the near future to start trying to connect them to implied therapeutics. The “in silico functionalization” bit is still perhaps questionable, but that’s a part of why I was wondering if this is on the radar for you all. Or perhaps you all know of much better ways to do this or pitfalls with the combination I’m proposing?

Maybe @ochoa and/or @MayaGhoussaini have done research in this direction already?

Hi Eric, we will definitely investigate the data sources you mentioned! Thank you!

In terms of variant pathogenicity etc., the Variant Effect Predictor (VEP) score for each variant is a feature in the l2g prediction model, so we use this evidence to predict putative target genes at associated loci. However, the model was not designed to infer LoF events, but it would definitely be an interesting extension to the pipeline!

Best wishes,