Johan Nakuci

Overall I find these approaches very interesting...My one concern is that no one checks if the reconstruction maintains the signal associated with the task or if it's only reconstructing the intrinsic signal. This is an easy test since you can do a GLM on the reconstructed signal.

Brain-like variability in convolutional neural networks reveals evidence-,uncertainty- and bias-driven decision-making Even when stimuli and tasks are held constant, brain activity fluctuates markedly across trials, yet it is not well understood how these fluctuations affect decision-making and behavior. Here we addre...

How does variability in activity impact decision-making?

Using CNNs, we show that trial-level variability reflects decisions with: (i) strong sensory evidence, (ii) low discriminability and (iii) sensory evidence provides conflicting information.

www.biorxiv.org/content/10.1...

Agree! and chemoarchitecture is a better predictor of brain connectivity than structure, at least the mesoscale .

A Practical Guide to Identifying Robust Clusters in Neuroimaging Data We outline practical strategies to enhance the reliability of clustering results in neuroimaging, addressing challenges in validation and interpretability. By integrating consensus-clustering, machin....

Human Brain Mapping | Neuroimaging Journal | Wiley Online Library onlinelibrary.wiley.com/doi/10.1002/...

Does a question like this even make sense for a system that is constantly receiving external inputs (eg. the brain)?

Lastly, we extend the framework to provide insight into the neuroreceptors underlying alterations in brain activity in Schizophrenia, Bipolar Disorder and ADHD

Moreover, we recover the binding profiles of LSD - 5-HT1a, 5-HT1b, 5-HT2a, and D2 - and Modafinil - D2, 5-HT1a, 5-HT2a, and NET -, demonstrating consistency with known pharmacological and neurobiological associations.

The neuroreceptors and transporters underlying spontaneous brain activity - Communications Biology PET-derived maps of 19 neuroreceptors accurately reconstruct BOLD signals, revealing two neuroreceptor modules orchestrating brain activity, recapitulate LSD and Modafinil binding profiles, and identi...

The brain has many neurotransmitter and neuromodulators, but which ones are driving brain activity at any moment? Using PET derived neuroreceptor maps we show that BOLD signal is driven by two neuroreceptors modules.

www.nature.com/articles/s42...

Mapping neurotransmitter systems to the structural and functional organization of the human neocortex - Nature Neuroscience Hansen et al. compile and share an atlas of neurotransmitter receptor/transporter densities in the human cortex and show that receptor achitecture reflects brain structure, function, dynamics, cogniti...

see www.nature.com/articles/s41...

In terms of how results can be biased (Marks blog), I agree with you completely. If the issue is inherent in the data, then it's difficult to see how an encoding model overcomes this. Overall, I see what you are trying to address.

In the simplest sense, I can fit a linear model to just Y ~ noise, and get a beta-value. However, that beta value is pointless without knowing how well the model fits the data. I could just be misinterpreting what you meant?

I agree with that but I am still not sure how the encoding model overcomes the issue? If noise > signal (x) then the model would be learning the noise. B/c without the decoding, then how do you know that what is being encoded reflects x?

I think I am missing something...are you saying that the noise level in the data is the issue or that if the image naturalistic (contains horse, barn and cat) then it is difficult to decode? My concern is that how can you then "determine" then the encoding model is reflecting the inputs?

Applying this framework, we identify distinct biophysical drivers of FC alterations in schizophrenia, bipolar disorder, and ADHD, offering insights into the unique characteristics of these disorders.

Our findings reveal that neuroreceptor congruence play a dominant role in shaping FC network features, while structural connectivity has a surprisingly minor influence.

Quantifying the influence of biophysical factors in shaping brain communication through remnant functional networks Abstract. Functional connectivity (FC) reflects brain-wide communication essential for cognition, yet the role of underlying biophysical factors in shaping FC remains unclear. We quantify the influenc...

We introduce a remnant functional networks-based framework to compare how functional connectivity is shaped by physical and molecular factors -- structural connectivity, physical distance, genetic and neuroreceptor congruence.

direct.mit.edu/netn/article...

Lastly, it's not my intension to make sweeping generalization about group-level analyses. (All typically developing individuals have a visual system, etc). However, it's hard to make nuanced points in 250 characters.

I am referring to brain-behavior or brain-pathology and etc relationships where individual heterogeneity is a "huge" factor. More importantly, to me group-level findings tend to be weak at best or simply noise at the worst case scenario. I know this is an old and on-going debate.

Let me rephrase my initial thoughts. I would prefer that we collect 1000 trials vs 1000 subjects (or longer resting-state, etc). I lean toward focusing on the the individual ("deep sampling" and etc). Group-level analyses have their place, but I am not sure how "meaningful" the results are.

Can you convince with besides the hypothetical "sure"? Overall, I think this kind of mechanism might work for language learning for instance in infants and/or for fine tuning in other instances, but in general I think it too slow.

Can learning via GD reach optimal weights sufficiently fast enough so that an organism doesn't get eaten (or fall off a cliff, literally)?

Brain signatures indexing variation in internal processing during perceptual decision-making Medical imaging; Cognitive neuroscience; Computer modeling

Also see, www.cell.com/iscience/ful...

By the way, the multiple patterns of activation are present in EEG too. See doi.org/10.1016/j.ne...

As we mentioned in the discussion section, if the subtypes reflected mind wandering or inattention, then this would presumably be reflected in the behavioral performance (RT), but the behaivoral performace was almost identical across subtypes.

To clarify, the subtypes are not driven by experimental factors (task conditions, etc). However, across the cortex, the brain-behavior correlation improves when subtypes are factored in. This is the "sensitvity" analysis (Figure 7). Overall, we can only speculate on what the patterns mean.

Different activation patterns are to be expected b/c Exp1 was numerosity task, whereas, Exp2 and Exp 3 were motion discrimination tasks. Sorry for the confusion, we didn't keep the subtype labels consistent across experiment, since the analysis was done independently on each experiment.

What does the mean mean? A simple test for neuroscience Author summary Neuronal activity is highly dynamic—our brain never responds to the same situation in exactly the same way. How do we extract information from such dynamic signals? The classical answer...

Also see this paper by Tlaie at al. at the single cell level. Putting it all together, these observations stem all levels of brain recordings and methods -- two photon, EEG, fMRI...

doi.org/10.1371/jour...

I started this idea/line of analysis in EEG b/c it's easier to "view" single trial brain responses in EEG.

Single trial variability in neural activity during a working memory task reveals multiple distinct information processing sequences Successful encoding, maintenance, and retrieval of information stored in working memory requires persistent coordination of activity among multiple br…

See doi.org/10.1016/j.ne... and www.cell.com/iscience/ful...

or noise in the data.