What do we mean when we talk about pattern separation of stimuli anyways?

By: Ryan Hunsaker

I have been having some interesting conversations on twitter with John Kubie (Link to his blog) concerning pattern separation and he has me completely flummoxed by a brilliantly simple question.

I realized his question was not pedantic as I had originally thought as I argued with him but rather was quite insightful and raised a philosophical point that is critical to understand what we mean when we say pattern separation. Since I feel this is an essential point we need to define as a field, I am throwing his question out to the group.

His deceptively simple question was : “What is a stimulus”?

Now, before answering, think about all that baggage that the term stimulus carries with it. Okay, now that we hopfully have all confused ourselves you can read on a bit further and I will explain my thinking on the topic.

Since I like to focus in on methodological questions, I chose to go down this road since this is an important methodological issue that needs a resolution. Also, the answer to this question seems to me to be the fundamental assumption upon which the rest of our research depends.

I used to have an easy answer for this question, but I have lost it of late. I still believe my old answer, but I am no longer confident using it as more than a working definition at best. Part of this reason is that Ray published a paper recently in Neurobiology of Learning and Memory that demonstrated that rats with dentate gyrus lesions result in an inability to remember a stimulus if it was provided in a spatial context, but did not result in a similar deficit if the stimulus was presented in front of blank space. In this case, it was a novel object recognition test is red or clear box. He was not the first to show the hippocampus can be involved for object recognition if there is a distracting context in rats, but the dentate gyrus lesion effect has a lot of theoretical implications for my work.

How I interpret these data is that the dentate gyrus is in a unique anatomical location to process large amounts of sensory information and to generate contextual representations (particularly spatial representations in rodents). If an object is presented is way that can be combined with the extramaze cues, then the object becomes part of the context and the rat cannot process and subsequently recall this representation without a dentate gyrus (i.e., a new object in that space will share the neural representation of the first object because of the larger context). If the object is presented in front of blank space, then there is no information for the dentate gyrus to generate a complex spatial representation/context, and the rat can use the normal cortically-mediated methods to recall the objects.

In a similar vein and more to the point of this post to spur discussion, how do we define a stimulus for our experiments? Can it be something that can be readily encoded in parts (for example a coat rack with a hat on it can be either encoded as a single entity or two items with a spatial relationship)? Does it clearly have to be a single item? Is it a problem if it can be given an easy to remember name? Can the item we test them on be lost in the background easily (figure/ground problem)? Is the test item actually a stimulus in itself or is it the result of a neural computation (Ray and I use the distance between objects as the manipulation, forcing spatial location to be the remembered stimulus)? Can it be a unitized representation of multiple stimuli? … And so on and so on.

I don’t know any of these answers. But I think this is one of the methodological questions we need to grapple with so we can start working toward cementing definitions and generating increasingly sensitive paradigms that test what we think we are testing.

Now to the tl;dr of this post, anyone want to jump into the trenches with me and venture an answer to John’s question? What exactly do we mean when we say, “stimulus”?

6 thoughts on “What do we mean when we talk about pattern separation of stimuli anyways?

  1. So after a bit of a break I came back to the site with another question, but I think I will just posit it here as it fits as a comment under my post.

    Is anyone interested in identifying, standardizing, and feeely posting under CC licensing actual sets of experimental stimuli for Human, primate, rat, and mouse experimentation.

    I am wondering this since I was casually reading a paper of Craig’s wherein they parameterized stimuli to generate functions of discrimination across stimuli. I was actually very pessimistic with regards to the human pattern separation tasks prior to this paper since I worried about the difference between degraded and partial cues and how pattern separation and completion would be preferred bases on the cue type.

    For rodents, I know Tim has done work with the touchscreen task development that is translatable across labs. But the feature ambiguity stimuli Ray and I have used in rats and mice have by and large not been parameterized to generate functions similar to those observed in the human research. I just wonder how important this is to standardizing results across labs in some way.

  2. I agree with Brad regarding the desire to think about input to the DG more carefully. As we mention in our review and in many conversations with Jim Knierim, understanding exactly how or whether the DG is decorrelating inputs cannot be done based on visual input or the stimuli themselves, but has to be done with consideration of the lateral and medial entorhinal inputs. Consistent also with Tim’s ideas I think that there is serial hierarchical decorrelation happening along the entire input stream with the conjunctive version of this decorrelation happening in the DG. This is also entirely consistent with Brad’s ideas that the DG is involved in the conjunctive encoding of what and where (and perhaps when), combining LEC and MEC inputs that way. Thinking about the type of information in addition to how the information is transformed is critically important in my opinion. We have done some work in this area recently with hi-res fMRI that is currently under review (given the exchanges on the website, I’m a little concerned we will receive “bipolar” reviews, as has been the experience for some here it seems).

  3. Along the lines of what Mike said, I think the stimulus is whatever information your experimental subject is exposed to. In my models, my stimulus is entirely controlled and known to me, but clearly that isn’t an option for people using biological systems. As a result, the best you can do is hold steady as much as you can and systematically vary only what you are interested in. I guess the two things that I would see as important are as follows:

    1) From a computational point of view, I think that the stimulus is everything the animal or model is taking in; not simply what is being varied. So if you’re dealing with a spatial displacement of a single object in a complex environment, the ratio of experimental stimulus to background stimulus – the “signal to noise” so to speak – is quite low. Even though we know the experiment, the rat/mouse (generally) doesn’t; so when inferring what the DG is doing, it is best to assume that it is dealing with everything. This can make it a really hard problem – what may be an easy separation in a vacuum may be a very challenging separation if 99% of the world is unchanged between two experiences.

    2) From a modeling perspective, I get frustrated when I see some studies (both in behavioral and in vivo phys) that seem to assume/imply that the differences in behavioral stimulus are somehow maintained as the differences in EC inputs to the DG (for instance). The DG is not decorrelating visual inputs or spatial locations (and related to the point above, is certainly not only decorrelating those experimentally manipulated stimuli). It is decorrelating the inputs from the medial and lateral EC. How differently positioned objects or rotated shapes are manifested in those EC inputs is completely unclear to me; and this makes me immediately wary of interpretations based on the initial parameterization. This may seem like a semantic difference, but I think it kind of underlies some of the communication challenges we’re facing. (Note: I love parametric behavior tasks, please don’t stop, I’m just voicing a desired consideration)

    1. I agree entirely with Brad on this one. Having recently done some work with the MEC and LECand processing distal space and proximal objects, it is tough. Particularly since the LEC = stimulus/object/proximal… (whatever we choose to label perirhinal inputs), and MEC = space, context, distal…(whatever we chose to label postrhinal/parahippocampal gyrus inputs) are not so clear cut. Ray and I found that the MEC and LEC do mostly what we say they do, but they also share some of the function of the other entorhinal areas.

      A few years back at the Winter Learning and Memory Conference in Utah Pierre Lavenex and Menno Witter took us all to task, quite loudly inf act, that we have a very annoying habit as experimentalists of oversimplifying our systems so that we can use an easy explanation. I think Brad (and to some degree Adam as well) are raising great questions that we need to find some way to address.

  4. Mike, that is exactly John’s point. I think we need to argue as a group how to best define “stimulus” in our studies and how we can avoid pitfalls or confounds associated with clumsy verbiage or definitions.

    To make it more complicated, I was talking with Ray and at times we end up actually testing “pattern separation” for a psychological construct (anxiety, affective value, distance between objects, spatial location). This totally changes what I mean by stimulus, since distance between objects for example is something calculated by the brain, and then we evaluate interference among different calculated differences (so comparing mental computations to each other).

    I know there isn’t an easy answer. In fact, I would love to hear computationalists chime in on this issue since stimulus selection is clear in their models but muddled in the behavioral experiments.

  5. I’ll venture a guess based on the origin of the word itself. In theory, a stimulus must stimulate or lead to an evoked response of some sort. It can be an item, a scene, a configuration, etc… or an electrical signal delivered by a stimulating electrode. I prefer a much more general definition for something like stimulus. What you’re referring to, in my opinion, is a finer distinction among different classes of stimuli, e.g. Object vs association vs configuration etc… which id entirely dependent on the experiment. Something could be an object in one setting but a member of a configuration in another as evidenced by Ray’s experiment and many others.

    Ultimately the stimulus is going to be everything that is available to the animal’s sensory systems but we define it narrowly in the context of the experiment. So for practical reasons one can say that an item is a stimulus or an air puff is a stimulus etc… Those are more operations definitions that are dependent on the experimental design.

    So to answer your original question what is a stimulus? I would say it all depends on your experiment. I don’t think there’s a good way to have a definition that transcends experimental design.

    My two cents.

Comments are closed.