The aim of this workshop called Bioinformatics and Artificial Intelligence (BAI) is to bring together active scholars and practitioners in the frontier of Artificial Intelligence (AI) and Bioinformatics. AI holds a tremendous repertoire of algorithms and methods that constitute the core of different topics of bioinformatics and computational biology research. BAI goals are two folds : How can AI techniques contribute to bioinformatics research ?, and How can bioinformatics research raise new fundamental questions in AI ? Contributions will clearly point out answers to one of these goals focusing on AI techniques as well as focusing on biological problems.

Important dates:

  • Abstract Submission Deadline: May 15th, 2017
  • Abstract and Paper Submission Deadline: May 29th, 2017
  • Author Notification: June 30th, 2017
  • Camera Ready Deadline: July 20th, 2017
  • Workshop: August 20th, 2017

Useful links:

logo Dr. Saman Halgamuge.
Professor and Director of Research School of Engineering at the Australian National University,
Canberra, Australia

Saman Halgamuge is a Professor and Director of Research School of Engineering. He is a Fellow of the Institute of Electrical and Electronics Engineers and a member of Australian Research Council (ARC) College of Experts for Engineering, Information and Computing Sciences. He was a Professor, Associate Professor and Reader and Senior Lecturer in the Department of Mechanical Engineering at The University of Melbourne (1997-2016). He contributed as Director (2013-16) of the PhD training centre Melbourne India Postgraduate Program (MIPP) of University of Melbourne, Associate Dean (2013-15) and Assistant Dean (2008-13) in International Engagement in the School of Engineering of University of Melbourne.

Title: Unsupervised Deep Learning: Applications in Metagenomics, Metabolomics and Drug Characterisation.
Abstract: (link for the full abstract with references)

Most of the existing Deep Learning methods rely on the assumption that all possible class labels sufficient to apply Supervised Learning are available. Although these types of learning algorithms can be generalized, their predictive power will be heavily constrained in the presence of partial information of a problem. For example, the classes that are available to a classifier are assumed to be ground truth, and their .correctness. is not generally questioned. In contrast to this approach, we propose a learning framework where the number of classes within a dataset do not need to be known a priori, and more specifically, the entire set of class labels are not required at the time of training. Instead, we propose to develop a method that will be able to infer the number of classes based only on the data and generate a more representative set of classes to train a robust classifier. Furthermore, we will also relax the assumption that these class labels are ground truth, and allow a degree of uncertainty in their correctness. An interesting solution for a subclass of these problems is Positive Unlabelled Learning [1].
Applying data analytics to microbial ecology has direct benefits to the design of vaccines and treatments to emerging pathogens, such as the Zika virus. In Metagenomic applications, very little may be known since we have only curated information pertaining to less than 2% of microbial diversity, and far less for novel variants of viruses. It is therefore not a realistic assumption that one can access all the true and underlying (organism) classes of any available data when analysing these organisms. Moreover, if we also consider the different and unknown number of effects that viral mutants can have on different hosts, and that these mutations could be linked to several environmental or geographical factors, we arrive at a complex, heterogeneous data set where labels are mostly unavailable, or any pre-existing labels available may be incorrect or not applicable to emerging viral strains. Even so, all these different types of data are essential to building a near-complete picture of the problem and understanding these pathogens at a deeper, more intimate level [2,3]. Statistical analysis of DNA sequence data has previously assisted us in identification of features that may further be used to discriminate species in a sample of multiple organisms using unsupervised learning methods [8-10]. Methods for increasing the resolution in realising the microbial population structure in a metagenomic sample is being worked on and coupling known data with unsupervised learning is found to be useful.
Repositioning of existing drugs as appropriate medication for previously not associated medical conditions can reduce the time, costs and risks of drug development by identifying new therapeutic effects. Investigating and understanding the interactions between drugs as well as how they work on our body is important in improving the effectiveness of clinical care. A method based on Positive Unlabelled Learning and Growing Self Organising Maps [1] is used on data available in DrugBank database. It was possible to infer 589 drug pairs that are likely to not interact with each other. Unsupervised Deep Learning also contributes in working with multielectrode array data [7].

logo Dr. Shoba Ranganathan.
Professor of Bioinformatics at the Deptartment of Chemistry and Biomolecular Sciences,
Macquarie University, Sydney, Australia

Shoba Ranganathan holds a Chair in Bioinformatics at Macquarie University since 2004. She has held research and academic positions in India, USA, Singapore and Australia as well as a consultancy in industry. She hosted the Macquarie Node of the ARC Centre of Excellence in Bioinformatics (2008-2013). She was elected the first Australian Board Director of the International Society for Computational Biology (ISCB; 2003-5); President of Asia-Pacific Bioinformatics Network (2005-2016) and Steering Committee Member (2007-12) of Bioinformatics Australia. She currently serves as Co-Chair of the Computational Mass Spectrometry (CompMS) initiative of the Human Proteome Organization (HuPO), ISCB and Metabolomics Society and as Board Director, APBioNet Ltd. Shoba's research addresses several key areas of bioinformatics to understand biological systems using computational approaches. Her group has achieved both experience and expertise in different aspects of computational biology, ranging from metabolites and small molecules to biochemical networks, pathway analysis and computational systems biology. She has authored as well as edited several books in Immunoinformatics as well as contributed several articles to the Encyclopedia of Systems Biology, published by Springer in 2013. She is currently Editor-in-Chief of Elsevier's Encyclopedia of Bioinformatics and Computational Biology as well as Editor - Bioinformatics of Elsevier's Reference module in Life Sciences.

Title: A protocol for finding missing proteins.

In the quest to uncover the entire human proteome, finding "missing proteins" remains the Holy Grail of scientists. In order to capture existing information, in addition to high-stringency MS data, we have launched the MissingProteinPedia (MPP; missingproteins.org), as an integrative biological database. While MPP incorporates automated data collection, novel tools for functional annotation and collated publications, there is an urgent need to identify a protocol for evaluating MPP data, to facilitate missing protein annotation jamborees. We will present how best to evaluate "extraordinary evidence" for missing proteins, with some exciting data confirming successful uncovering of some missing proteins.