Firstly, thank you for your tireless work maintaining this incredible resource.
I work on classification tasks in drug discovery and OT is one of my underlying data sources.
Could you please help me understand your conceptual framework regarding non-associations between targets and diseases / adverse events?
As you likely know, in classification both positive and negative examples are required. However my intuition is that it is unrealistic to declare definitive non-associations between a target and diesease or adverse event, given the difficulty of exhaustively and reproducibly demonstrating that.
It would be very helpful to have your reasoning in your words.
This complex question might depend on the use case you want to answer. The Platform does not interpret evidence as “negative” when defining target-disease associations. For example, benign mutations reported in ClinVar have a score of 0 because they are considered neutral (not negative). Even clinical studies that stopped due to a negative reason (efficacy) are not regarded as negative evidence because you could consider that there was enough body of evidence to start a clinical study, and the negative outcome (e.g. lack of efficacy) might be the attributed to the drug molecule not to the gene-disease association. When addressing is this gene causal for this disease, there is very little evidence that can truly tell you some gene is not associated.
In the locus-to-gene assignment, we had to go through the same process of defining negatives, and we concluded that we should use the absence of evidence for not-functionally related genes (STRINGdb) as a negative. There is always a risk that this is not entirely true, but it’s the best approximation for the given task.
In the target prioritisation view, evidence can be regarded as favourable or unfavourable. I wouldn’t consider unfavourable targets (e.g. presence of safety events associated with the target) as negative evidence, but they denote some negative connotations when asking for specific questions (e.g. safety).