ML is fairly common in genomics, but for identifying predictive variables for cancer status, it's difficult. The training set is a matrix where rows are people (where some have cancer and some don't) and columns are genomic features (mutation, methylation, etc). You can easily have hundreds of thousands of features but getting even a thousand cancer patients enrolled in a study and sequenced is expensive and slow.
So, even though there are many "AI in biotech" companies out there, for predicting cancer status, most eventually end up hand crafting a small number of features based on extensive knowledge of cancer biology. The ML model tends to be simple and far less important than the features.
2. Liquid Biopsy -- many companies are working on diagnostic tests to detect cancer early, monitor treatment, detect recurrence, etc. Some of these products are already available but most are in R&D phase.
3. Immuno-Oncology -- the most promising development in cancer treatment in a long time. Successful tumors evade the immune system by essentially hiding themselves. Many companies are working on treatments to assist immune system in different ways. Some products have been in the market for a while now and have been pretty successful.
I also have a difficult time reading large numbers without commas. Because I use Alfred (mac command launcher) quite a bit, I wrote up a simple workflow to re-display selected numbers with commas. Not as universal as the font-based solution but might be useful to some:
> For to accurately and repeatably describe positions
IMO, for most use cases, you don't need to describe positions; you provide something analogous to a SQL where clause.
Most querying in cell biology appears to content-based rather than coordinate-based. "drive down this artery 1000 cells" is not useful because your target might be 1010 cells down next week. Instead, you might say "drive down this artery until you find a cell with this particular receptor on the cell wall." Rather than keeping track of how many cells you've passed, you randomly bounce around the artery until you bind to a compatible receptor.
Actually, a 'debug' or 'dry-run' mode that simply prints out the git commands without running them would be useful for all kinds of users. Newbies can learn from it and pros can make sure that the commands are going to do what they expect it to. With git, I know what each command does but my concern with legit (and similar) is that the higher level commands might not do exactly what I think based on their name/description.
But random-mutation is just ONE part of Darwin's theory. The other critical part is natural selection and that is what allows for organisms that are specialized for their specific environments. Epigenetics is certainly important but it does not mean that evolutionary theory is incorrect.
I would guess that any observed differences are from lifestyle (food type, amount of sun, exercise, beauty products, hair styles) rather than epigenetic differences. I haven't seen any papers where the phenotype of epigenetic changes is as clearly observable as "facial structure and skin".
Epigenetics is a really exciting area of Biology right now as studies begin to investigate the causes and consequences of DNA methylation. It seems that, after all these years, Lamarck may be due for some credit!
It should be noted, however, that this is not the only memory mechanism available to the cell. Gene expression, microRNA expression and protein modification can all act as switches (and thus memory) that place the cell in some discrete state. They're fairly transient and non-inheritable though.
Racial disparity doesn't just result from racism or "like preferring like". There are many institutional and structural issues that are partially causal and a frank discussion could help uncover them as they relate to the startup world. Sadly, frank discussion of race is very difficult in this country.
Btw, I do agree that the valley is very meritocratic so I don't think racism is a big factor.
Space might be an issue but it's also strategic. I need the higher-resolution screens and have bought (and will continue to buy) the 15" versions even though I don't really need the better CPU or GPU. I'm sure there are many others like me who grudgingly or not will fork over the extra cash.
I remember people asking the same kinds of questions when Wolfram Alpha was released. Alpha and Watson work in very constrained spaces and transferring their success to general search is very much non-trivial. So, no Watson will not make a dent in Google.. It'll be a very helpful tool in specialized areas but scaling it out is not a matter of a year or two..
I agree but my conclusion based on this is that, for the foreseeable future, we'll need hybrid systems that can merge expert knowledge with machine learning.
In the coming weeks there will be a lot of hand-wringing about Watson's performance on Jeopardy and what it means for humanity and jobs and so on.. I think it's important to remember that the last mile in many tasks will still require humans and that AI should be seen more as human augmentation rather than replacement.
The Personal Genome Project [personalgenomes.org] is aiming to recruit 100,000 people to publicly release their DNA sequence and medical data. The website currently has phenotype and medical history data and genotyping data for the first ten participants who are all well-known scientist.
I would say "Brain-inspired". As I see it, Numenta's model (as of about a year ago) is based on (1) the hierarchal organization of neurons, (2) the presence of feedback loops in neural architectures and (3) the importance of temporal processing even for static scenes. This doesn't include any intracellular details nor any of the larger and/or specialized brain structures.
So, even though there are many "AI in biotech" companies out there, for predicting cancer status, most eventually end up hand crafting a small number of features based on extensive knowledge of cancer biology. The ML model tends to be simple and far less important than the features.