Differential Gene Expression - fcrimins/fcrimins.github.io GitHub Wiki

On the PBS NewsHour last night (10/21/15) they profiled a family in Toronto who had 4 children--all boys--and one of them had autism. They researchers are comparing the genomes of people with autism to close family members of people without autism and they say they're close to finding genes that are indicators for the disease. Their goal is to either sequence 11k people's genomes or the genomes of 11k people with autism and their close family members.

Adaptive Gene Expression (10/8/15)

  • Are all networks learned? The network in the brain is learned. A convolutional neural net (CNN) is learned.
  • And if so, what does it mean to "understand the brain"? If it's anything similar to "understanding a CNN" then there doesn't seem to be much value in understanding how it does what it does as opposed to understanding how it learns what it learns.
  • And what if gene expression is similar? What if gene expression is a learned network? It certainly changes throughout life and in response to things like smoking; it's adapting--i.e. learning. And if that's the case it might be silly--to some extent--to search for what genes do as opposed to how they learn what to do.
  • Of course some proteins do very specific things, but maybe others do similar things and can pick up the slack for each other. Protein A might accomplish task X in some people, while protein B handles X in others. Sickle Cell Anemia is caused by a very specific gene mutation, but it expresses differently in different people (according to Inquiring Minds episode 56 with Steven Johnson). Perhaps that's because it's "learned" differently in different people.
  • What we do know is that the genome is a network of many interactions. It seems like such networks must be learned--rather than programmed. Are all such networks learned? Evolution did much of this learning, but that doesn't account for changes in gene expressions over time.

Email: "The 'what did I eat toilet'"

  • Reconstruct what someone ate from their poo. This could help to reduce the number of variables in genetics studies (GWAS), one of the main problems of which is that they have too many variables.
  • Also are someone's changes in gene expression levels during such a study ever indicative of anything? Is anything like that ever measured? If you can tell if someone's a smoker by their expression levels then surely, at some horizon, changes in those levels have meaning. Use machine learning to figure it out. But first you need inputs, e.g. poo.

The price to pay for focusing on one thing changing, is not seeing all of the things changing

  • change blindness
  • [https://www.youtube.com/watch?v=PlhFWT7vAEw&list=PLE6Wd9FR--EfW8dtjAuPoTuPcqmOV53Fu&index=16]
  • FWC - The same happens in data modeling. When we're focused on modeling the state, it would be much easier to focus on modeling changes in state.