adredish

adredish, 4 months ago

@elduvelle @albertcardona

Adobe Acrobat does OCR. Just go get the hard copy, take some pictures with your phone and voila. PDF.

adredish, 5 months ago

@elduvelle

IME, Mastodon is definitely more siloed. In my experience, the twitter algorithm (pre-X) was actually shockingly good at finding stuff I wanted to see but didn't know about. (Importantly, I was very strict about not doing any politics on the site. I got my political news elsewhere.) I [a] didn't officially follow anyone, [b] only "liked" science and work-related posts, and [c] only responded to science and work-related posts. But I found that the work that I discovered on Twitter was much broader than Mastodon, where I have to follow individuals to try to create that algorithm.

adredish, 5 months ago

@manisha @elduvelle Thanks for the reply.

How does one follow hashtags and what are guppe groups?

adredish, 5 months ago

@NicoleCRust @thetransmitter

But we are making progress with the weather! Weather predictions are vastly more reliable than they were a few decades ago. Even if the brain is chaotic, that doesn't mean it's not understandable. And it doesn't mean it's not manipulable (a word I like better than "controllable").

It may mean we need to think about controlling it on different levels.

Moreover, we control brains all the time. As Bill Newsome has been saying in his public talks --- There are two ways to get him to get up and run out of the room. You can either manipulate all of the neural firing and molecular states to move his muscles in the right way. Or you can tell him "your wife's on the phone and it's an emergency." Both of these are ways of controlling brains.

adredish, 5 months ago

@PhiloNeuroScie @marcwhoward Do they also recommend increases in science funding and NIH R01 grant sizes?

adredish, 5 months ago

@marcwhoward @PhiloNeuroScie

The danger about fewer PhDs is that it hurts diversity. One factor (not the only factor) that leads to a lack of diversity is that, because of the way that the US is economically structured, many people from marginalized communities have had limited opportunities (for research experience, for quality educational experience, etc). Limiting the number of PhD slots means that it is less possible to take risks on people who have either had limited opportunities or who have stumbled sometime in their lives. I feel (very strongly) that what we need to do is make it possible for MORE people to take the risk of being a PhD student and provide better safety nets if it doesn't work out.

I'm not sure how much this translates to limited opportunities if we have fewer postdocs, but fewer postdocs definitely means less work done. I'm just saying that higher salaries with the same total budget means less work done.

I'm all for higher postdoc and graduate student (and postbac and technician and staff scientist) salaries - I just think we need to increase NIH budgets to match. There is a real problem with academia in general where we take budget cuts and say "our work is so important, we'll find a way to do it anyway".

Some data: We were spending 12% of GDP on science in 1965. We now spend 3%. The real moonshot cost $300 billion in 2020 dollars. (Compare the brain initiative $5b, cancer "moonshot" $2b).

#AcademicChatter #PostDocs

adredish, 5 months ago

@LorenAmelang @elduvelle

Yes! DUA! [Don't use Acronyms!]

I once struggled to understand a grant that was doing these very weird motor experiments trying to get at Head Direction (HD) cells and Hippocampus (HC) in humans. Until halfway into the grant, I realized these were Huntington's Disease (HD) patients being compared to Healthy Controls (HC) at which point the grant made MUCH more sense. 🤣​🤣​🤣​

adredish, 5 months ago

@danielmk @elduvelle

Let's not change how we individuals "enjoy writing grants". Let's create a system that does not require writing grants - or at least minimizes the difficulty. It's a shame that @Nature is using its bully-pulpit to scold individuals instead of to argue for better systems.

Distributing money by the illusion of "best science as written in a grant" is provably an inefficient, wasteful, local minimum that hurts the scientific enterprise.

adredish, 5 months ago

@elduvelle @eLife

Wait. What is this step 4 (send to another journal)? Why would you do that? If you do that, #eLife is just a fancy #bioRxiv. I thought the whole point was that the "version of record" in step 3 was the final published version and you listed eLife as the journal.

Is this wrong?

adredish, 5 months ago

@elduvelle What I would do is stop at the #eLife step but include the keywords they boldface in a subline below the entry when I list the paper in the CV. (For many promotion documents, one needs to have these sublines anyway explaining your role in the paper.) I think people will accept the sublines.

adredish, 5 months ago

@elduvelle YMMV but in my experience, eLife and NNsci historically have had similar impact on grant-getting and promotions. The real question is whether people are changing their opinion of #eLife with the new system. (I haven't seen it, but it is too early to tell.)

adredish, 5 months ago

@tdverstynen @NicoleCRust @albertcardona

I wish other fields (Calcium imaging comes to mind! but also electrophysiology... and head-fixed behavior... and everything else...) were as explicit about their limitations as the fMRI imagers generally are.

This is why we need THEORY. Theory is a means of bringing together the constraints from a variety of sources to create understanding.

Honestly, IMO, the biggest problem right now is the attempt to find the "perfect experiment" instead of trying to fit one's little experiment into the literature.

Now, if only we could convince scientists to actually read that literature...

adredish, 6 months ago

@albertcardona

Thanks for these citations! I like that we can now stop talking about "the #connectome". A single #connectome doesn't exist. Anymore than "a single #genome" does. We need to start seeing species as populations with variability (this is how #evolution works! - by shifting those population distributions around).

As a practical note (coming back from #SFN2023), I'm getting worried about all of the experiments that are taking some specific mouse strain, making some observation (specific connections from a small subset of cells in structure A project to this specific part of structure B), and generalizing from that to all mice (and all rats and all humans). I wonder how much the various very specific connections that are being traced with viral techniques are specific to that "genetic family" of mice and how much variation we should be expecting.

I think we are vastly underestimating the variability in these life forms we are studying.

adredish, 6 months ago

@albertcardona

Maybe not in the c elegans field (I wouldn't know), but the fMRI field talks of "the #connectome" ALL the time.

adredish, 6 months ago

@albertcardona @BorisBarbour @deevybee @PubPeer

How does pre-registration get to this question? Or are you answering a different thread?

Please check back and re-read the post that started this thread. I'm saying something very different about social media and perception and the creation of community.

adredish, 6 months ago

@neuralreckoning @UlrikeHahn @johnntowse @albertcardona

Here's some real statistics. From NINDS.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8397261/

This is a fantastic paper and well-worth the read.

One does not need a K99 to get a faculty position. In fact, one does not need prior funding to get a faculty position.

One does not need CNS papers to get a faculty position.

Lots of people go through intermediate positions (non-traditional tracks) to get to faculty positions.

adredish, 6 months ago

@BorisBarbour @albertcardona

I think too much is made of data sharing data as a means of identifying dodgy stuff. It is very rare that one can access the full raw data (and no one is going to re-preprocess the data) and the truth is most neuroscience replicates beautifully. In my observation, the neuroscience debates (at least in the worlds I live in of behavioral neuroscience), the arguments are ones of interpretation, not fact.

If the only point of data sharing is to catch dodgy stuff, it would be much more cost effective to hire an army of people to replicate experiments. That would actually catch the dodgy stuff. Although that runs into problems of "what is replication?". (is a different species replication? is a different strain? is a different cohort with different prior experience?)

The problem is that sharing data well is a headache. It's expensive both in terms of time, money, and effort. It requires a large amount of additional information that was not used in the original experiment (because now the data might be used for something else and we better be sure it's valid for that something else). And it's not rewarded. Thus, there is no incentive to share data. Making it mandatory is just yet another regulatory mandate that makes it harder to do science.

Which would you rather I do? Share my data? Run a larger n? Do a second follow-up experiment? You get to pick one. I'm not at all sure that "sharing data" is the most important thing for science.

Do we think that catching a little bit of dodgy stuff (even assuming data sharing would do that) is worth the cost of everyone having to share their data?

I suppose you could give me a sixth year on my R01 if I share my data. That would definitely incentivize me to do so. But that's the scale I would need to incentivize me to share data well.

PS. We do share our data. All of our recent papers have the data uploaded to data share servers, but it is just the data we used to make the paper. It is not "shared well" to enable people to do lots of stuff with it.

Edit: I saw that this was part of a longer chain where data was requested and not provided. It looks like this may be due to data restrictions. (I don't know if this applies here, but it can be very tricky to share human data because it has to be de-identified, which is not always so easy.)

adredish, 6 months ago

@BorisBarbour @albertcardona

Thanks for the discussion. An interesting blog. Some responses.

1. The most valuable and concrete products of the investment are the discoveries that come from it. Science is about discoveries that allow us to see things in new ways. I deeply disagree with the data/code worship that is in the current zeitgeist. (I realize that is an iconoclastic view, but it is one I hold.) Data and code are merely steps along that way. In my view, the products are discoveries that allow us better control of the world because they have identified a better understanding of mechanism and process.

Data and code are useful, sure. But they are not the the goal, nor the "product" of science.

2. I definitely do agree that IN THEORY data sharing improves quality, efficiency, and integrity. But in practice, I haven't seen it.

Particularly, in the realm of efficiency: More often than not, data reuse is misused - it attempts to ask questions for which the experiment was not well designed. And, yes, data sharing improves quality and integrity, but it is a very very expensive thing to do right.

Following on that point, I think it is extremely disingenuous to simply demand that data be shared without addressing the costs. The number of regulatory burdens that we now face to do science because of a few (and I do believe they are very few) bad apples is making it harder and harder to do the science.

3. I absolutely agree that it is hard to fabricate a full real fake data set. Maybe I'm naive, but I don't think most of my colleagues are fabricating data. So the question I ask is how much we are willing to spend on checking each others' data vs getting the opportunity to explore and try new experiments.

4. I'm totally OK with saying that everyone has to put up the data
   that they are using for the paper in some central place (and we do, for the same reason that we fill out IACUC and IRB forms, because regulations require it), but we shouldn't expect that data set to be useful or easily interpretable.

And I stand by my statement that this is mostly unnecessary and just makes the science more expensive and move more slowly.

There's a fascinating discussion about these issues by Adam Mastroiani arguing that science is a "strong link" problem not a "weak link" problem. https://www.experimental-history.com/p/science-is-a-strong-link-problem

If he's right (and I think he is), we should be encouraging more opportunities for exploration.

Now, as we move from discovery to engineering to implementation, we move from a strong link to a weak-link realm. So, an RCT for some drug that is going to be brought to market and given to millions of people should absolutely be pre-registered, data-made-available, the whole nine yards. But for the initial science of it, maybe we need to reduce the burden on doing experiments.

adredish, 6 months ago

@BorisBarbour @neuralreckoning @albertcardona

I do think we agree here. I am very OK with "sharing data as is".

I am also very OK with "learning how to format it better with time". (Actually, I like that a lot. It's a very different sentiment than I've usually heard. I'm going to start using that argument about data sharing.)

I am a little worried about data reuse misuse, but given the argument about "strong link", if someone misuses my data without checking with me, that's on them, not me. It should all be detectable in the wash.

Some servers (like OSF) just let you put files up there as is. Other servers (like Dryad) are pains-in-the-a** about making the data accessible. (We fought with Dryad for a while about our data formats for a recent eLife paper - providing Matlab code to load them wasn't good enough because "Matlab is not free software". So we ended up putting the data on OSF.)

So, in the end, I think we do agree. Put up the data as is, include what code you've used to generate your figures (again, as is). And over time, we'll learn how to format things better.