Improve the Results section

[praeducer]

• [ ] Make purpose of why we are sharing the results more clear. Show how theater studies are kind of demos of the value of Vaults. Show reliability and reproducibility benefits.
• [ ] Cite other research more clearly.
• [ ] Show how we can reproduce
• [ ] Show how this contributes to open science
• [ ] Reference COBRE data studies and data sets. Show results from analysis on COBRE data. This should demonstrate reproducibility.

This is similar to the smoking study Kelly and Harsha worked on.

[praeducer]

To clarify, are we focusing on reproducibility of the results but not the actual scientific findings? What do we use from the other papers we published?

From Vince:

• Show we can get the same results with the same data.
• Show we can get other data and combine it with the Vault data.
• Show we can run computations while users are offline since Vaults can always be online.

For one of those studies, do we have an example of processing multiple nodes. Make sure that is highlighted.

From Javier:

• It helps to kick the tires. Try out COINSTAC with ease.
• Removes barriers to performing this kind of research. Faster path to getting research done.

Make sure to cite this paper and address it's concerns:

4.2.4 Alternative approaches

Some research initiatives and databases have attempted to develop an alternative approach to sharing human subject data to better protect subjects' privacy or to respect subjects' autonomy on how their data should be used. For example, instead of sharing raw individual-level data, the Enhancing Neuroimaging Genetics through Meta-analysis (ENIGMA) provides analysis protocols to run at a local site and perform meta-analysis on results (e.g., summary statistics) returned from the local sites (Thompson et al., 2014; Thompson et al., 2020). The Collaborative Informatics and Neuroimaging Suite Toolkit for Anonymous Computation (COINSTAC) is another large-scale research consortium that offers tools for decentralized analysis which allows researchers to conduct both meta- and mega-analysis without pooling the data at one central place (Ming et al., 2017; Plis et al., 2016). Groups of users run common analyses on their local sites using their own data and the results of these analyses are then synchronized to the cloud and undergo aggregate analyses processes using all local data. The federated computing of COINSTAC also provides heightened privacy protection using advanced statistical algorithms, such as differential privacy, and this model may open up a way to gain access to data that researchers are unable to share due to local regulatory restrictions (Ming et al., 2017; White et al., 2020). However, it is important to note that federated computing could make data sharing more burdensome because it would require significant synchronized agreement efforts among all the sites to run the analysis locally or to change the protocol.

https://onlinelibrary.wiley.com/doi/full/10.1002/hbm.25803

[praeducer]

Brad and Kelly discussed clarifying what we want to say with the results in the results section. They will get input from Sunitha and Vince Some things we might want to demonstrate: COINSTAC Vaults can get the same result as centralized data. What vaults we have created Researchers with their own datasets can increase the sample size for their analyses Vaults make studies more reproducible Demonstrate the kind of analyses that will appeal to neuroscience researchers. ICA, VBM

[praeducer]

@sunithabasodi Can you confirm how you coded for sex and diagnoses? For example, trying to understand if males or females have more grey matter. We're trying to interpret the images. Kelly is concerned some values may be inversed. Some edits may need to be made for clarity, but mostly Kelly needs to know to write to it in the results better.

[praeducer]

@spanta28 Do you know how we could reference results from an analysis on COBRE data to show the reproducibility of studies using Vaults?

[praeducer]

Sandeep's notes:

Hey team, one change here needed on page 9, "Fig. 7 Rendered images show voxel-wise β values corresponding to the age, sex and diagnosis covariates using MCIC sMRI data in COINSTAC. Both Vaults were then combined in a regression analysis to examine diagnostic effects while accounting for age and sex." Instead of Both Vaults, it should be COBRE VBM data Vault combined with local MCIC data.