Data Science and Sharing Team

National Institute Of Mental Health

Investigators

Linked publications, trials & patents

Abstract

For our fifth annual report we have provided some highlights of our team's activities over the past fiscal year. Data Sharing Our team continues to collaborate with Joyce Chung and the NIMH Clinical Director's Office on the Healthy Research Volunteers Protocol (NCT03304665). Phenotypic data for 1,358 participants have been deposited into the OpenNeuro data repository (https://openneuro.org/datasets/ds003504) using the Brain Imaging Data Structure (BIDS) standard. 149 of these subjects also have standardized MRI scans and 63 have MEG scans. These data will be made publicly available once the associated data paper has been accepted for publication. DSST staff have also provided assistance in making data sets standard compliant and publicly available for Audrey Thurm's and Ben White's groups. Data Curation With the continued growth of large public datasets, data curation/management continues to be our team's most requested service. To manage these demands, we have brought new pools of storage online including one petabyte of shared solid-state disk (SSD) space on the NIH High Performance Compute Cluster (HPC) which is available to all NIMH intramural researchers. Here we provide access to over 100,000 MRI-based scan sessions across 31 different shared datasets. We maintain a comprehensive list of these datasets on our website (http://cmn.nimh.nih.gov/dsst). In collaboration with Dr. Armin Raznahan's group, we are also now curating Genotype-Tissue Expression (GTEx) datasets from the NHGRI Genomic Data Science Analysis, Visualization, and Informatics Lab-space (AnVIL). This 100 TB dataset is now accessible from the NIH HPC available for use by the wider NIMH intramural program. Training The DSST is continually providing ad hoc training consulting with researchers and trainees throughout the NIH intramural program, however three more structured training offerings are detailed below: - Three of our staff (Dustin Moraczewski, Jessica Dafflon, and Arshitha Basavaraj) participated in the ABCD ReproNim (https://www.abcd-repronim.org) as both teaching assistants and high-level students who could provide guidance and mentorship to more junior researchers. The course had over 100 students enrolled from around the world. The course provides detailed instruction on reproducible methods, gold-standard, NIH-funded share data (ABCD), and larger principles of Open Science. - In July, Eric Earl served as an instructor for NeuroHackademy (https://neurohackademy.org) a selective and highly respected training course that is now in its fifth year. Graduates of Neurohackademy frequently go on to become leaders in the reproducible neuroscience community. - In August, Adam Thomas led a team at the NIH Summer Internship Programs Codeathon where trainees at various levels learned and practiced data science and reproducible computing. Preregistered Replication Starting in September of 2019, our team mentored a one-year post baccalaureate student, Nik Goyal, who submitted a paper to the Royal Society of Open Science on their preregistered replication track. In this publication model, only the study design and methods are peer reviewed before acceptance. If the design and methods are sound, the publication of the study is ensured regardless of the results. This model combats publication bias by ensuring that negative results are published. Nik's submission was enthusiastically accepted and the analysis demonstrating the finding is indeed reproducible in an independent dataset has been completed. It is now under final review for compliance with the pre-registration which is available on the Open Science Framework (OSF) repository (https://osf.io/8ctqz). Data standards and sharing for Positron Emission Tomography (PET) In collaboration with Dr. Bob Innis, our team is working to advance and encourage data sharing in the PET community. This has entailed the creation of standard nomenclature, a new BIDS specification for formatting and sharing PET data, and a new public repository for sharing standard-compliant data. For more details on this project, see the NIMH annual report entitled OpenNeuroPET. Collaborations The DSST continues to collaborate with investigators on projects and publications that involve large shared datasets. In the past year, these collaborations have included multiple projects with our sister group, the Machine Learning Team (MLT). As applying machine learning methods to answer neuroscientific questions has become more prevalent, we seek to improve interpretability when applying deep neural networks to predict brain state. See our preprint for more information: https://arxiv.org/abs/2004.11114. In another project with the MLT, we aim to leverage information across multiple large datasets to improve upon individual phenotypic prediction models utilizing multiple imaging modalities. In our collaboration with fellow IRP researcher Arman Raznahan and the Section on Developmental Neurogenomics, we assisted with processing the structural brain images from almost 40,000 subjects within the UK BioBank dataset in order to examine how variations in the X-chromosome relates to neuroanatomical variability. This work was recently published in Nature Neuroscience (https://doi.org/10.1038/s41593-021-00890-w). COVID-19 Our work on COVID-19 survey data continues in collaboration with Drs. Francisco Pereira, Joyce Chung, and Melissa Brotman. Carl Harris, who has recently joined our group as a post bac intramural trainee has taken the lead on this project. He has made substantial progress with machine learning models designed to predict longitudinal outcomes using kernel regression techniques. Data collection and analyses are ongoing.