Is a computational neuroscience MD/PhD feasible?

[deleted728364]

I'm currently working in a "lab" (independent PI who doesn't train PhDs/postdocs) where we use computational methods to develop diagnostic tools for neurological disorders. This is not really a "proper" comp neuro lab, as we don't really use these methods to study the actual function of brain regions. Instead we use machine learning as a sort of black box (rather than model of a neural circuit).

I told my current PI about my interest in machine learning + neuroscience and he had some reservations:

1: He thinks that the machine learning trend in neuroscience is a bit of a bubble and the use of ML has already been exhausted. Also, collecting useful data is the main bottleneck, and that's ultimately a question of (scarce) funding.
2: He's not sure that a PhD in comp neuro would be the best for an MD/PhD. Instead, I should focus on a biological aspect of the brain/some disease and use computational methods as a tool if they're useful.

Regarding 2, a comp neuro MD/PhD is feasible, right? One clearly applicable area I’ve thought of is adaptive deep brain stimulation. Plenty of neurologists/neurosurgeons who use computational stuff like ML for DBS. Some work with engineers/neuroscientists, but to me it seems like there’s enough overlap. Perhaps one of the biggest examples is Jamie Henderson, who works with Krishna Shenoy on BCIs. That said, there are many neurologists who work with computational scientists/engineers/neuroscientists work at in comp neuro. Another examples are these two MD/PhDs from UIC.

Or is computational neuroscience still a bit too disparate to anything in medicine for a physician scientist? Is it an area better suited for computer science/statistics or other basic science/engineering folks?

Some other example labs w/o MDs I can think of (but may not prove my point): 1, 2, 3, 4

I feel like this sort of work is what the MD/PhD degree is intended for - clinicians who want to do basic science rsrch.

---

Members don't see this ad.

 

[tr]

Not my field but this sounds totally normal to me, there are definitely people who do MD/PhD in computational neuroscience. Columbia has a strong group (or did), with Larry Abbott, Ning Qian, maybe others. I think you can also do MD at UCSB and PhD at Salk with Terry Sejnowski. It's out there.

I disagree that you should pick a brain region or disease of interest now. You can do that later. Computational skills are transferable and a great thing to acquire in PhD while you are young and flexible.

Obviously you will need data to work with, you could leech off a large brain mapping consortium project or join a neurophysiology lab that collects their own. My impression is more that current tools have led to an oversupply of data and an undersupply of people who have the tools for analysis, but maybe that's something more specific to my area and experience.

 

[dl2dp2]

Your supervisor appears not well informed.

PhD in computational neuroscience (and in a similar vein, biomedical informatics/computational biology, etc.) is a different ball game vs. other biomed PhD program. This group of PhDs have very transferable skills that in general represent a significant value add to an MD degree. It's feasible to get a job that pays about as much or even more than a physician with a PhD in this set of disciplines and appropriate subject matter training and work experience (mainly in ML, but also in some other subfields, such as pure informatics). This leads to a shortage of people of a certain background in academia, which leads to 1) higher demand; 2) easier grants; 3) more collaborative opportunities.

The actual content of the science matters very little. Few people eventually do whatever they got trained on during PhD.

 

[deleted728364]

The actual content of the science matters very little. Few people eventually do whatever they got trained on during PhD.

His argument exactly - he says that the ML bubble will burst, or ML will become so common place that a PhD in the subject isn't really that useful. His phrase: "parkinson's/epilepsy/etc will always be here". As in, doing graduate work in these areas isn't a waste because no one's gonna cure thesis diseases quite soon. Computer modelling, however, has seen many different trends (simplex way back in the day, finite element modelling, and now machine learning).

So I'm better served by getting a PhD in some neurobiology related thing. Basically - the "dataset" with which I'll use ML stuff with. His argument is that domain knowledge (and any ML-practitioner will agree) is actually more important if you want to make a useful algorithm. (more important that knowing the fine details of implementing algorithms.

I can see his point, but my solution was to do a PhD co-advised by a neurobiology dude and an ML dude. This way I avoid the risk of getting an "obsolete" PhD, even if that's not gonna be a risk in the first place.

 

[dl2dp2]

I can see his point, but my solution was to do a PhD co-advised by a neurobiology dude and an ML dude. This way I avoid the risk of getting an "obsolete" PhD, even if that's not gonna be a risk in the first place.

It's also useful to remember that when you take career advice from people, it's good to keep in the back of your head whether the advice has some ulterior objective. It's useful to ask advice from a lot of different people to assess for commonalities and notice differences.

 

[ExcitatorySynapse]

I am at a smaller program and there's a guy doing exactly that kind of PhD (computational methods to study neurological disorders) in my program, and I very much hope to do something similar as well.

I'm not exactly sure how your envisioned co-supervision would go. I know in some "disease focused" labs you generate data and then analyze it (both from models and human samples). IME, such labs use fairly established genomic/proteomic/transcriptomics methods and they are less interested in developing new methods by themselves. Initially I was interested in such work but I honestly don't think I want to touch a pipette again and I feel that learning how to develop methods in ML/statistics while I have 3-4 years to do so will better position me to "direct" my own research going forward.

That being said, I don't think you will lose out by focusing on data analysis, IMO you necessarily have to be knowledgeable about the biological contexts from which you derive your datasets from, the pathophysiology of the diseases that you are working on. Unless you do a degree in the stats/CS department, I don't think you will lose the biological context. Plus, it's not impossible to develop a knowledgebase on a different disease through reading. Having known plenty of straight CS/stats people, they're actually allergic to biology, but you obviously wouldn't be the same.

As the other posters have stated, specializing on a disease now doesn't seem like it has that much value. After PhD, you have 2 years of MS + several years of residency. Interests change and people jump from different areas all the time in their postdoctoral work. Someone I know went from GI -> Viral Oncology (straight PhD but still). I know a NSG w/ a lab who did his PhD on optics and now does work on deep brain stimulation w/ animal models. I think its easier to learn/develop wet lab skills or knowledge about a disease than it is to develop highly technical computational skills post-PhD.

 

[deleted728364]

I looked through CMU/Pitt's MSTP trainees that are doing comp neuro stuff, and several of them have this arrangement. To me it makes intuitive sense, but I bet it's also dependent on the culture of each advisor's department/if the advisors are willing to or have collaborated in the past.

I honestly don't think I want to touch a pipette again

Felt this. Hated microbio lab with a passion.

 

[kchan99]

I think its easier to learn/develop wet lab skills or knowledge about a disease than it is to develop highly technical computational skills post-PhD.

For learning wet lab vs. compututional skills, it depends on interest and time. Either can be learned during a postdoc after residency.

 

[StilgarMD]

For learning wet lab vs. compututional skills, it depends on interest and time. Either can be learned during a postdoc after residency.

As someone trying to make the transition from wet lab stuff to computational skills, I actually think on this point a lot. I'm nearing the end of my PhD, and there is a great sense of pressure that unless I learn the Comp stuff during my PhD, It won't happen. I just can't imagine being a resident, sitting down with a calculus or stats text, really digging into things to understand what is appropriate to use and when, at the same time as I have any clinical responsibilities, a wife, and probably a kid or two. I can barely get through the to-do lists I have now, it seems like such lofty things are a "now or never" thing. Though I could be wrong.

 

[kchan99]

As someone trying to make the transition from wet lab stuff to computational skills, I don't actually think on this point a lot. I'm nearing the end of my PhD, and there is a great sense of pressure that unless I learn the Comp stuff during my PhD, It won't happen. I just can't imagine being a resident, sitting down with a calculus or stats text, really digging into things to understand what is appropriate to use and when, at the same time as I have any clinical responsibilities, a wife, and probably a kid or two. I can't barely get through the to-do lists I have now, it seems like such lofty things are a "now or never" thing. Though I could be wrong.

I did wet lab before and during med school. I did computational work and a little bit of wet lab during fellowship.

 

[ExcitatorySynapse]

For learning wet lab vs. compututional skills, it depends on interest and time. Either can be learned during a postdoc after residency.

I definitely agree that you *can* learn anything but it's been my impression that a postdoc level, people usually expect you to have the knowledgebase and less likely to invest time in you learning a skillset when they could conceivably hire someone who already has it. I'm happy if it's otherwise true tbh.

 

[kchan99]

I definitely agree that you *can* learn anything but it's been my impression that a postdoc level, people usually expect you to have the knowledgebase and less likely to invest time in you learning a skillset when they could conceivably hire someone who already has it. I'm happy if it's otherwise true tbh.

In all honesty, I had previous coding experience in high school, which I had forgotten, and a significant interest in computers.

 

[StilgarMD]

I did wet lab before and during med school. I did computational work and a little bit of wet lab during fellowship.

What field do you work in? Its is heartening to know such a thing may be possible.

 

[spoopydoopy]

Bit of a negative take, but it's also seemed harder to me for people to pick up "good computational habits" as they move on in their training. Simple yet crucial skills like commenting etiquette, version control, just general clean programming - it all seems easier to learn early on. Anyone can learn to code, but are you going to learn to elegantly and efficiently run a computational project as a resident or postdoc?

OP, comp neuro is a great PhD route if you find a supportive MD/PhD program and good mentor. I would really question anyone arguing with the utility of ML in modern medicine. I have a whole separate rant about wishing more people felt free to get weird with their PhD.

 

[kchan99]

What field do you work in? Its is heartening to know such a thing may be possible.

I'm in neurology clinically and MRI analysis for research, but previously I was in cancer biology. I also worked with tissue from time to time. Hours wise, I worked as hard and as long during my postdoc as I did during residency.

With my cancer biology background, I was able to dabble a little in whole exome sequencing analysis.

Bit of a negative take, but it's also seemed harder to me for people to pick up "good computational habits" as they move on in their training. Simple yet crucial skills like commenting etiquette, version control, just general clean programming - it all seems easier to learn early on. Anyone can learn to code, but are you going to learn to elegantly and efficiently run a computational project as a resident or postdoc?

I'm not in computational neuroscience. My work has been in MRI so the number of languages to learn is limited, and I don't distribute code. Everything I write is for my own work, and often I use programs written by others. In other words, my computational work is not hard core.

It won't be possible to do serious computational work during residency but during postdoc is possible depending on length and what you put into it.

 

[happyhoneybadger]

Come to electrophysiology lab