TDbrain Challenge

The goal is to develop clinically actionable biomarkers, that should generalize to ‘unseen’ data, so we have marked 30% of clinical and treatment outcome data from the TDBRAIN database as REPLICATION. These are blinded for prospective validation and replication purposes. Interested researchers can share their diagnostic or treatment-outcome predictions on these REPLICATION data with us (This email address is being protected from spambots. You need JavaScript enabled to view it.), so we can perform a ‘blinded out-of-sample’ validation for them, and post the results on this page, so that reviewers for journals can also consult this independent validation of their algorithms. Conversely, reviewers or editors faced with ‘too good to be true’ claims, can refer researchers to this resource, to perform validations and share them with us for verification and publication. We hope this will accelerate the discovery and validation of clinically actionable biomarkers that we hope will soon benefit patients and improve psychiatric and neurological care in the clinic.

For more information about the TDBRAIN dataset also read our published data descriptor here in Nature Scientific Data, and for access to TDBRAIN, click here. Note that the subjects marked 'REPLICATION' are mainly intended for treatment prediction outcomes (e.g. response/non-response, remission/non-remission). For prediction of 'diagnostic' status we have a seperate held-out set that can be downloaded called 'Adult validation sample'.

For treatment prediction biomarker we are using nPPV (normalised Positive Predictive Value; nPPV>110% considered succesful) as primary outcome. For diagnostic biomarkers we use Balanced Accuarcy as the primary outcome (and should exceed >60% to be considered succesful). The table below is sorted with best results on top. Our focus is on ADHD and MDD, so OCD results are omitted.

Also feel free to suggest other predictions, sich as age- and sex-prediction or other metrics.

Click on the image.