Two scientists may look at the same data and draw different conclusions. Faced with a problem to solve they may see different solutions as the obvious way to go. The cause of this is scientific taste: one's crystalized collection of priors about how the slice of nature of interest works.

I like to think of taste as tinted glasses: you can look at a phenomenon through different lenses and notice different things with each. These views are not to be thought of as right or wrong in isolation. Rather, they may be better or worse suited for a particular purpose in a given context.

One such view is that of genetics: In its purest form, this view is interested with genes as functional units. So one would be looking at gene knockouts and knock ins, editing genes, measuring gene expression, looking at correlations between gene variants and traits (GWAS). In the face of a problem to solve or a disease to cure, one would reach for a genetic explanation, which gene(s) are responsible for the problem. One would think of DNA as a 'code' that determines everything downstream. Therapeutically and scientifically, this view lends itself to the search for genes to edit and better tools to do those edits.

This view is very common in biology. Genes are indeed really important. In an earlier post I pointed at turnover rates as a heuristic to determine the causal power of something in biology, and the genome, as something that's relatively stable, is the most causally important entity in organisms.

In many cases it is a very useful view: It is the view that brought us PCSK9 inhibitors to give a simple but powerful example. And all else equal, drugs developed against target that have genetic support are most likely to work.

It is, however, not the only view nor the most useful one in every situation.

If you study aging unavoidably one ends up looking at the whole organism and there one sees the obvious truths that

  • All cells (neurons, hepatocytes, etc) have the same genome yet behave very differently
  • Cells from aged and young individuals behave differently yet they also share most of their genome
  • You can wipe out the aged phenotype from cells without touching their genomes
  • Most disease, even in carriers of disease risk variants, develops with age; even familial AD patients take at least two decades to get the disease

It goes deeper than that:

  • Nuclei of murine cancer cells, if placed in an oocyte have their cancer phenotype erased, giving rise to seemingly healthy mice (Mintz et al. 1975). If cancer was purely an outcome of having a mutation, this shouldn't be possible. These results have been replicated a couple of times (Utikal et al. 2009; Kim et al. 2015). These animals derived entirely from cancer cells have a higher propensity to develop cancer: the effects of the mutations eventually return, but for a while cells are able to behave 'as they should' as part of a multicellular organism.
  • Fibrosis or scars broadly are not just collagen deposited: they turn over. As a result one can ask what keep that phenotype (or rather collection of phenotypes) in place (an imbalance of fibroblast and macrophage activity; Adler et al. 2020). This can be leveraged therapeutically (Miyara et al. 2025; Gao et al. 2024) by altering the signaling between the cells or removing the overactive celltype altogether.
  • One can fight cancer by being aware that hitting it hard can lead to evolved resistance. Instead, one can dose lower amounts of different kinds of treatments at different times, so that one keeps the tumor in check without it evolving resistance to the overall regime (West et al. 2020; Zhang et al. 2023; Zhang et al. 2022)
  • Cellular senescence, prematurely termed 'irreversible' can be reversed if one tries to understand what keeps senescent cells that way (An et al. 2020)
    • One might ask: Why not just do a CRISPR screen? People have done that (Wang et al. 2021; Li et al. 2024; Li et al.2023, Liu et al. 2019; Jing et al. 2023), but somehow the mechanism described in An et al. seems to have eluded them. Note that these CRISPR screen all show different hits. This happened perhaps because with small molecule screening one can hit a gene some other members of the same family for a more potent effect whereas CRISPR screens might be too precise, so cellular redundancies are able to work around it.
  • You can stop "Alzheimer's", at least in some mouse models, by ablating all microglia after the disease starts (Spangenberg et al. 2016)
  • Children can regenerate their fingertips when cut (Schultz et al. 2018)
  • Cancers, regardless of their mutations, posess a different electric charge in their cell membrane compared to non-cancer cells (Chen et al. 2016; Yang et al. 2013). This phenotype can be leveraged to develop therapeutics that target all cancer cells (Chang et al. 2019)
    • A related approach, using electric fields to slow cancer progression is already FDA-approved and in use for glioblastoma; there are other companies pursuing this eg this one
    • This is in stark opposition to precision oncology, which tries to develop therapeutics targeted to specific mutations; cell membrane potential is an emergent phenotype, not something driven by a single gene.

These things are not so surprising from a systems perspective. I suspect this view is less common because it essentially says that we can understand biology leveraging higher level abstractions (like interactions between cells or circuits of genes), but these abstractions are not the crisp and clean ones we are used to from the formal sciences, rather they are nebulous, but that's okay. They don't have to be crisp, they have to be good enough.

The wiring diagram in that An et al. paper is far from complete, but it was useful in their context for their purpose of interest. The standard view it's either that we can understand things in a very local context under tightly defined conditions (knocking out a gene) or we have to kind of give up (or leave it to AI to solve, maybe). Admittedly the experiments to test systemic approaches tend to be more complex: Testing in vivo, testing multiple cell types combined in vitro (organoids), or testing multiple drugs at the same time are harder. But additionally, if one really wants to tell a very defined causal story, that's easier to do with the genetic approach. Whereas in the studies earlier where the cancer phenotype is abrogated, we know it is but we don't know the specifics step-by-step about why: we don't have a mechanism.

That is perhaps another way to see these two views: the genetic view limits the scope of what one can think about but in exchange you get really good explainability. The systems view lets you go further, thinking of novel experiments or hypotheses to try, in exchange for less clarity or certainty in the story.

If you want to get a taste for this perspective, I recommend reading Systems Medicine, from Uri Alon or watching this from Michael Levin.