Does anyone know how close are we at the moment from being able to say "we're getting X smell, so here are the genes producing X producers"? I know we can (most of the time) say, here are the sites producing a specific protein. But are we able to find sites for second and higher order results yet?
It's luck of the draw. We understand perhaps 15-20% of the genes/proteins in the human genome fairly well. We have fairly general notions of what the rest are but lack specific understanding. We also discover novel functions for genes/proteins we thought we knew well.
One way to discover function is to what's called a "genetic screen": Assuming one develops a robust way of measuring those chemicals; we fix that as the "phenotype" we're looking for. We can then disable each gene (with CRISPR, or RNAi, or chemical mutagenesis..) and see if the chemicals get produced (or stop getting produced); and then sift through those. This strategy is far from being exhausted; and people keep coming up with cool phenotypes to measure.
One idea I've been excited for some time is using ML to robustly annotate behaviors from mice by using wide-range microphones and 24/7 video as the phenotype; and use a combinatorial genetic screen to rigorously discover genetic contributors to behavior (aggression, mobility, curiosity, socializing behavior etc.) The ML algorithm can be initially trained by hand-annotated behaviors (annotated by post-docs etc.)
