2^nd Machine Learning in Immunology Competition

(webpage: http://bio.dfci.harvard.edu/DFRMLI/HTML/natural.php)

We are pleased to announce the 2^nd Machine Learning in Immunology (MLI) competition. Machine learning applications in immunology play a key role in the field of immunoinformatics. This competition follows the very successful competition sponsored by ICANN09.

Experimental studies of immune system and related applications such as vaccine design, optimization of therapies, or other combinatorially complex applications. These experimental approaches are time-consuming and expensive. New approaches involve pre-screening by computational models, followed by experimental validation using selected key experiments.

Peptides that bind Major Histocompatibility Complex Molecules (MHC) are important targets in studies of cell-mediated immunity, regulation of immune responses, vaccine research, and transplant rejection. A number of computational models have been developed for prediction of MHC-binding peptides (MHC ligands) and standards for their assessments have been developed (see Zhang et al, 2011). MHC-binding peptides are normally determined using biochemical binding assays. Only a subset (approximately 10-20%) of MHC-binding peptides that are naturally processed are considered as useful targets for vaccine development. In silico identification of candidate MHC-binders is a standard methodology in screening targets of immune responses and vaccine targets. Some of algorithms for in silico prediction of are highly accurate, while others need improvement. Algorithms that can accurately identify naturally-processed MHC-binding peptides would reduce cost of target validation by 5-10 fold. Machine learning-based methods offer a great promise for further advancement of prediction systems in this field.

Participants in this competition will develop prediction methods for in silico screening of naturally processed MHC-ligands. The training data will be provided by the organizers and are also available from public databases, such as IEDB, SYFPEITHI, MHCPEP, ANTIJEN, and MHCBN. Organized data sets are available at the DFRMLI site. The competitors will be presented with new sets of experimentally identified target data and asked to perform predictions. The groups that demonstrate best predictive performance on these carefully selected data sets will receive the InCoB/ICIW MLI competition prize and certificates. The predictive performance will be assessed using criteria similar to those used in 1^st MLI competition.

This competition will help identify machine learning methods that will improve upon currently available tools for prediction of peptide binding to MHC molecules. In particular we invite participants from mainstream machine learning community to join the competition and help address this challenging problem, traditionally done by bioinformaticians.

Organizers
April 2012

Task Description
The target of competition will be prediction of peptide binding to a subset of naturally processed MHC class I molecules. The list of training peptides will be provided on July 25. The list of test peptides will be revealed on August 1.

Participation
The competition is open to any individual except for the organizers and their direct employees. The participants will remain anonymous unless they request for non-anonymous participation or if they are among the winners. Participants are not required to attend the Workshop, but are encouraged to do so. See Rules for further details (to be modified).

a) Conditions of participation
The Machine Learning in Immunology competition (MLI Competition) is open to any person who accepts the MLI Competition rules. The competition is a part of the InCoB 2012 conference that will be held in Bangkok, Thailand on October 3-5, 2012 (see http://www.incob2012.org). The competition will be concluded at the 2^nd Machine Learning in Immunology Workshop on Oct 5. Competitors are not required to attend the MLI Workshop, but are encouraged to do so. Each individual can register only once for the competition.

b) Restrictions

• The organizers of the competition are excluded from participation as competitors.
• The organizers reserve the right to restrict the number of competitors from the same institution or a collaborative group.

c) Anonymity
The competitors are required to register, but they can opt to remain anonymous and participate by using a pseudonym. The registration will be via the Web page (to be announced later). Participants can pre-register by email to Vladimir Brusic (vladimir_brusic{at}dfci.harvard.edu). All public communication will contain only pseudonyms of participants who opt to remain anonymous. The names of the winners will be revealed in the MLI Workshop.

d) Data
Training data will be posted on the DFRLMI web site on July 25. The prediction targets will be posted on the same site on August 1. Prediction results must be submitted to the organizers by September 20 (Midnight GMT). The participants who submitted the prediction before September 20 will be able to resubmit predictions until September 25. The maximal number of submissions will be 3, and the latest submission will be used as the competing entry.

e) Prediction target molecules
The predictions will be revealed later. Training data sets will be available at DFRMLI site (http://bio.dfci.harvard.edu/DFRMLI), but the participants can use any additional data sets (for example IEDB, SYFPEITHI, literature, etc). For competition, 6-8 target data sets will be modeled.

f) Evaluation
Each competitor will submit one entry per target set. Each prediction will be ranked and the overall best ranked submission will receive the first prize. The best individual performing predictions (for individual predictions) will receive the certificates. The Evaluation Committee will reserve the discretion in making the final decision.

g) Prizes
The MLI competition grand prize will be awarded to the winner. The certificates will be awarded to the winners in each of the 6-8 categories.

h) Publication
Selected entries, deemed appropriate for the immunological audience, will be invited to submit for publication in the Journal of Immunological Methods.