This article has been Reviewed by the following groups
Drug repurposing can accelerate drug development while reducing the cost and risk of toxicity typically associated with de novo drug design. Several disorders lacking pharmacological solutions and exhibiting poor results in clinical trials - such as Alzheimer’s disease (AD) - could benefit from a cost-effective approach to finding new therapeutics. We previously developed a neural network model, Z-LaP Tracker, capable of quantifying behaviors in zebrafish larvae relevant to cognitive function, including activity, reactivity, swimming patterns, and optomotor response in the presence of visual and acoustic stimuli. Using this model, we performed a high-throughput screening of FDA-approved drugs to identify compounds that affect zebrafish larval behavior in a manner consistent with the distinct behavior induced by calcineurin inhibitors. Cyclosporine (CsA) and other calcineurin inhibitors have garnered interest for their potential role in the prevention of AD. We generated behavioral profiles suitable for cluster analysis, through which we identified 64 candidate therapeutics for neurodegenerative disorders.
Article activity feed
This gives rise to some questions about the reproducibility of the“missing” compounds - whether they truly display a CsA-like behavioral profile.
I appreciate the context and transparency in this section! I wonder if comparing and clustering the previous results together with the results generated here could highlight both areas of methodological improvement and increase confidence in the biological effects of your compounds of interest.
64 compoundsdisplaying CsA-like behavioral paradigms (Table 1)
The way you list out these compounds really helps me understand and interpret the results! It's such a diverse list, are there any compounds that surprise you? Conversely, it seems to me that some of the compounds you found are to pathways with strong behavioral effects generally (e.g. the neuromodulatory pathways involving dopamine, norepinephrine, and serotonin). Are there ways to account for general motor effects when identify CsA-like compounds?
When compared against DMSO-vehicle controls, 89% ofthe CsA-like compounds induced statistically significant changes in behavior.
This seems like an important caveat. Would it be possible to remove compounds without an identifiable change in behavior before clustering? It's good that they are a minority (~10%) but but seems odd to call compounds CsA-like based on behavioral effects if it's not clear they have a behavioral effect
Pearson pairwise correlations of CsA and the 47 CsA-like compoundsidentified by both K-means and hierarchical clustering.
This is a nice way to visualize the data, it really helps me get a sense of which compounds are similar to Cyclosporin A and which aren't. Is it possible to do this for all the compounds rather than just the ones found by clustering? Similarly, how to you interpret compounds identified by clustering but with low (e.g. 0.04) correlation to CsA?
The resulting behavioral profiles consisted of 25 behaviors
These behaviors seem interesting but somewhat arbitrary. Is there a way to analyze the data you collected in an unbiased way to extract the most relevant features?
We found that CsA clusters with 58 other compounds (Fig 3)
Could you specifically point out where CsA lies here, I'm having trouble understanding the results without that reference point. Similarly, could you point out which point(s) are DMSO, is it all the unlabeled data points? Knowing where the control points lie would be very helpful!
We identified the optimal number of clusters (k = 4) using the elbow.CC-BY 4.0 International licenseavailable under a(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is madeThe copyright holder for this preprintthis version posted September 13, 2023.;https://doi.org/10.1101/2023.09.12.557235doi:bioRxiv preprint
It would be awesome to see the curve you used here and some examples of other clustering results. Particularly because clustering is fundamental to the downstream analysis some more information on the effects of clustering differently would give the results more context
Intotal, we examined 50,496 larvae: 42,048 larvae treated with small-molecule compounds,and 8,448 DMSO-treated larvae
This is an impressive screen! Particularly since you have so many controls, could you show more analysis of baseline variability in larvae behavior and how it changes with DMSO? It would be helpful to understand the typical range of larval responses to interpret the experimental compound-induced changes.