The DREAM Challenges are crowd-sourcing to solve complex biomedical research questions.
Together, we share a vision to enable individuals and groups to collaborate openly so that the 'wisdom of the crowd' provides the greatest impact on science and human health. Over sixty crowd-sourced DREAM Challenges have benchmarked informatic algorithms in biomedicine. DREAM has had over 30,000 cross-disciplinary participants from around the world that have volunteered as solvers. Over 105 academic journal publications have resulted from DREAM Challenges covering a range of disease areas.
Learn more about the DREAM process - Pose > Prepare > Engage > Evaluate > Share
More information and Submission Details regarding presenting at RSGDREAM 2022 Coming Soon.
To develop predictive models for risk of preterm birth in pregnant individuals based on vaginal microbiome data.
There is a controversy whether the vaginal microbiome of pregnant patients differs between patients who deliver term and those who deliver preterm. (1–4) Our challenge seeks to use crowd-sourcing to explore such differences and determine if computational models using vaginal microbiome data will enable improved prediction of preterm birth in order to benefit those who suffer from this condition.
Globally, about 11% of infants every year are born preterm, defined as birth prior to 37 weeks of gestation, totaling nearly 15 million births.(5) In addition to the emotional and financial toll on families, preterm births have higher rates of neonatal death, nearly 1 million deaths each year, and long-term health consequences for some children. Infants born preterm are at risk for a variety of adverse outcomes, such as respiratory illnesses, cerebral palsy, infections, and blindness, with infants born very preterm (i.e., before 32 weeks) at increased risk of these conditions.(6)
The ability to accurately predict which women are at a higher risk for preterm birth would help healthcare providers to treat in a timely manner those at higher risk of delivering preterm. Currently available treatments for pregnant women at risk of preterm delivery include corticosteroids for fetal maturation and magnesium sulfate provided prior to 32 weeks to prevent cerebral palsy.(7)
There are several factors known to be associated with PTB, including maternal age, body mass index (BMI), education, smoking, history of PTB, a short cervix, and genetic polymorphisms.(8–11) Nevertheless, there are currently no clinical tools that enable the early and reliable prediction of preterm birth.(12)
There is some indication that the vaginal microbiome plays a significant role in adverse pregnancy outcomes, specifically preterm birth. Previous studies have shown that there are significant differences between the vaginal microbiome of patients who deliver at term and those who deliver prematurely. Vaginal microbiomes with increased diversity as well as communities where Lactobacillus is not dominant have been associated with PTB.(13–15) We hypothesize that this data could be used as a potential avenue for predicting which women are at a higher risk delivering birth.
Patients and healthcare systems alike would benefit from the formation of a precise determination of risk for PTB as outlined in the Challenge.
Decoding how gene expression is regulated is critical to understanding disease. Regulatory DNA is decoded by the cell in a process termed “cis-regulatory logic”, where proteins called Transcription Factors (TFs) bind to specific DNA sequences within the genome and work together to produce as output a level of gene expression for downstream adjacent genes. This process is exceedingly complex to model as a large number of parameters is needed to fully describe the process (see Rationale, de Boer et al. 2020; Zeitingler J. 2020).
Having the ability to understand cis-regulatory logic in the human genome is an important goal and would provide insight into the origins of many diseases. However, learning models from human data is challenging due to limitations in the diversity of sequences present within the human genome (e.g. extensive repetitive DNA), the vast number of cell types that differ in how they interpret regulatory DNA, limited reporter assay data, and substantial technical biases present in many omic methods. To overcome these issues, we have recently created high-throughput measurements of the cis-regulatory activity of millions of randomly generated promoters in the single-cell organism Yeast (de Boer et al. 2020). Here, the expression level generated by each promoter sequence is measured via a fluorescent reporter gene regulated by a promoter (Sharon et al. 2012). The set of randomly generated promoter sequences is so large that it rivals the complexity of the entire human genome, which gives us unprecedented power to learn the many parameters required to understand gene regulation (see Rationale). Because both human and Yeast cis-regulatory logic uses similar principles, we hope that the model architectures learned on yeast data can inform on how to create models for the human genome.
In this competition, the participants will be given expression measurements of millions of randomly generated promoter sequences to train machine learning models that predict gene expression from sequences. The participants will be provided with TPU Research Cloud resources to help train their models.
Brain tumors are among the deadliest types of cancer. Specifically, glioblastoma, and diffuse astrocytic glioma with molecular features of glioblastoma (WHO Grade 4 astrocytoma), are the most common and aggressive malignant primary tumor of the central nervous system in adults, with extreme intrinsic heterogeneity in appearance, shape, and histology, with a median survival of approximately 15 months. Brain tumors in general are challenging to diagnose, hard to treat and inherently resistant to conventional therapy because of the challenges in delivering drugs to the brain, as well as the inherent high heterogeneity of these tumors in their radiographic, morphologic, and molecular landscapes. Years of extensive research to improve diagnosis, characterization, and treatment have decreased mortality rates in the U.S by 7% over the past 30 years. Although modest, these research innovations have not translated to improvements in survival for adults and children in low- and middle-income countries (LMICs), particularly in sub-Saharan African (SSA) populations.
The Brain Tumor Segmentation (BraTS) Continuous Challenge seeks to identify the current, state-of-the-art segmentation algorithms for brain diffuse glioma patients and their sub-regions. Training and blinded validation data are made available year-round for building and evaluating segmentation algorithms. Evaluation metrics for predictions on the blinded validation data are returned immediately to participants with a continuously updated leaderboard. This continuous challenge will culminate in an annual, state-of-the-field evaluation where we will ask participants to submit containerized versions of their best models which will be run against a held out testing data set, and the results presented at the annual MICCAI conference.
The BraTS training and validation data available for download and methodological development by the participating teams describe a total of 5,880 MRI scans from 1,470 brain diffuse glioma patients and are identical to the data curated for the RSNA-ASNR-MICCAI BraTS 2021 Challenge. The unseen hold-out testing data will include the BraTS 2021 Challenge test data, as well as new data from out-of-sample sources including i) an independent multi-institutional dataset covering underrepresented SSA adult patient populations of brain diffuse glioma (Africa-BraTS), and ii) from another independent pediatric population of diffuse intrinsic pontine glioma (DIPG) patients. These new out-of-sample datasets are meant to evaluate the generalizability of existing models to new, out-of-sample populations. All challenge data are routine clinically-acquired, multi-institutional multi-parametric magnetic resonance imaging (mpMRI) scans of brain tumor patients.
While durable responses and prolonged survival have been demonstrated in some lung cancer patients treated with immuno-oncology (I-O) anti-PD-1 therapy, there remains a need to improve the ability to predict which patients are more likely to receive benefit from treatment with I-O. The goal of this challenge is to leverage clinical and biomarker data to develop predictive models of response to I-O therapy in lung cancer and ultimately gain insights that may facilitate potential novel monotherapies or combinations with I-O.
Immuno-oncology (I-O) therapy targeting the PD-1 pathway has transformed the treatment landscape in advanced non-small cell lung cancer (NSCLC), with the combination of I-O with or without chemotherapy being the current standard of care in the first-line setting for those patients who are ineligible for targeted therapy.1-5 While durable responses and prolonged survival have been demonstrated in some patients treated with I-O, there remains a high disease burden and a need to improve the ability to predict which patients are more likely to receive benefit from treatment with I-O.
The Bristol Myers Squibb-Sage Bionetworks Anti–PD-1 Response Prediction DREAM Challenge is the first DREAM initiative and collaboration in the I-O space. Like other DREAM Challenges, the Anti–PD-1 Challenge is a crowdsourced effort that looks to advance our understanding of foundational questions in biomedicine through open-science collaboration. We invite experts and innovators in genomics, computational biology, and translational biomarker development to participate in this Challenge that aims to identify predictive biomarkers for I-O therapy in lung cancer. The deidentified, validation dataset for this Challenge comes from an international, randomized, open-label phase 3 trial (CheckMate 026) of anti-PD-1 nivolumab vs platinum-based chemotherapy in patients with previously untreated advanced NSCLC.6 By exploring RNA-sequencing data for predictive signals of efficacy and resistance to nivolumab, the Anti–PD-1 Challenge seeks to improve our ability to appropriately select patients most likely to benefit from I-O treatment and to gain insights that may facilitate potential novel monotherapies or combinations with I-O.