DDBJ Data Analysis Challenge

"DDBJ Data Analysis Challenge" is a machine learning competition using "'International Nucleotide Sequence Database' data", which is one of the life science big data provided by DDBJ. Participants need to submit their generated machine learning models to a collaborative website, UnivOfBigData. Even college students and/or researchers outside of life science field, can get an opportunity of studying "Machine Learning and Data Mining", through this Challenge. And, to join this Challenge more easily, DDBJ provides NIG Supercomputer System as your computer resources.

DDBJ Data Analysis Challenge 2016

Date Content
Jun 27, 2016 Start Accepting Applications:
NIG Supercomputer System application
NIG Supercomputer System OSS Installation request
Jul 6, 2016 Start Date
Aug 21, 2016 Deadline for Applications:
NIG Supercomputer System application
NIG Supercomputer System OSS Installation request
Aug 31, 2016 End Date
Sep 30, 2016 Result

Challenge Task

DNA Data Bank of Japan (DDBJ) supports a big data resource called by DDBJ Sequence Read Archive (DDBJ SRA), which contains DNA sequences genearated from high-throughput DNA sequencers. The secondary analytical database, ChIP-Atlas database (Dr.Oki of Kyushu Univ.) provides the annotation data of chromatin feature regions on genome sequences.

At this challenge task, please predict whether genomic regions corresponding to input DNA sequences includes chromatin feature regions. Chromatin feature region is related to on-off function of gene expression, and corresponds to peak regions on a genome sequence of the ChIP-Atlas database.

Challenge's target species is a plant. The number of conditions for the target plant is often over 100. The number of conditions in the challenge is reduced and composed of eight conditions for saving time of try and error on data modelling.

Input training data: 60,000 DNA sequence
Input test data: 10,000 DNA sequence
Output training data: 8 conditions correct answer sets

One input sequence is composed of 200 bases, that is a ACGT sequence fragment with 200 length,
where the sequence is encoded as 01 code [Example: AATGC ... = 10001000000100100100 ...] so that
the length of a sequence is 800 digits.
Corresponding code: A = 1000, C = 0100, G = 0010, T = 0001, Other exceptions = 0000

Output correct answer sets of 8 conditions is also encoded as 01 code.
True answer is one, which means that the input DNA sequence contains chromatin feature regions.
Likewise zero is false answer so that it does not include the chromatin feature region.

On the submit stage, please submit the probability of true prediction with 10,000 rows (test axis) and 8 columns (condition axis) in BigData University website.


1st Prize of DDBJ Challenge Awards 2016 Information and Mathematical Science and Bioinformatics Co., Ltd.
2nd Prize of DDBJ Challenge Awards 2016 RIKEN ACCC Bioinformatics Research Unit
MATSUMOTO Hirotaka(representative*), OZAKI Haruka(*)
*They participated in this Challenge as a team.
3rd Prize of DDBJ Challenge Awards 2016 BITS Co., Ltd.
OKAYAMA Toshitsugu
Student Prize of DDBJ Challenge Awards 2016 Master's Degree Program 1, Graduate School of Information Science and Technology, The University of Tokyo
KATO Takuya


AUCModel DesignTool Version
1st Prize 0.94564 *2 Classifiers(Extremely Randomized Trees, CNN)
*Ensemble Learning(Stacking)
*External Data(Genomic Position, Gene Structure Annotation)
2nd Prize 0.89859 *2 Classifiers(CNN, Product of Genomic Distance Decay Parameter and  Nearest Training Data Output)
*Ensemble Learning(Averaged)
*External Data(Genomic Position)
3rd Prize 0.85428 *7 Classifiers(Naive Bayes for Multivariate Bernoulli Models, Logistic Regression, Random Forest, Gradient Boosting, Extremely Randomized Trees, eXtreme Gradient Boosting, CNN)
*Ensemble Learning (Stacking)
Student Prize 0.84318 *3 Classifiers(LeNet like CNN, DeepBind like CNN, Variable filter DeepBind like CNN)
*Ensemble Learning(Soft Voting)

DDBJ Challenge Committee

DDBJ Challenge Committee
Eli Kaminuma, PhD : Center for Information Biology, National Institute of Genetics, Assistant Professor
Hisashi Kashima, PhD : Department of Intelligence Science and Technology, Kyoto University, Professor
Toshihisa Takagi, PhD : Center for Information Biology, National Institute of Genetics, Professor

DDBJ Data Analysis Challenge has been approved ethical review by NIG Institutional Review Board (IRB).