Find out how ICT can support biomedical and clinical researchFind out more. From Clever cars to clever farms... Embedded Systems
Highlights

1.   Methodology for classification of anti-learnable data

  • Some recent experiments with bio-mining of genomic microarrays for prediction of response to treatment and ECG records of abnormal heart conditions have shown that these datasets have unusual properties. They manifested a phenomenon named “anti-learning,” where many machine learning procedures systematically generate models (almost) perfectly classifying training set but almost perfectly misclassifying an independent test set.
  • Theoretical explanations of this counter-intuitive phenomenon have been developed. In particular, they link anti-learning to a signature of hidden competition for limited resources, which is biologically plausible.
  • A number of dedicated algorithms have been developed and demonstrated to classify successfully both the “anti-learnable” and normal “learnable” datasets.
  • Prediction of response to cancer treatment from gene expressions in tissue biopsy for oesophageal adeno-carcinoma is a real life diagnostic task with strongly anti-learnable signature. Attempts to explain and solve this prediction task has given a direct stimulus and motivation for this research.

2.   Marker selection methods for a novel Carcinoma of Unknown Primary pathology test  

 

  • In 5-15% of new cancer cases, a tumour's site of origin cannot be accurately determined by conventional pathology methods leading to suboptimal treatment.
  • Jointly with Peter MacCallum Cancer Centre and Circadian Technologies, NICTA is developing a clinical test based on Quantitative Polymerase Chain Reaction (QPCR) expression profiles.
  • The developed methods have demonstrated an accurate predictor allowing an individual to be profiled on a single 384 plate QPCR plate. Further techniques are being developed to allow prediction of two patients from a single QPCR plate.

 

3.   Algorithms and software tools for detection of cancer DNA micro-aberrations

 

  • Novel statistical signal processing combined with high resolution SNP microarrays allows for up to 1000-fold improvement in the resolution of cancer DNA aberrations. The ability to identify these mutations will assist our understanding of the pathogenesis of cancer, particularly if they are consistent across multiple tumour samples. Narrow regions of change in the human genome often go undetected as algorithms tend to regard individual outlying points as noise and exclude them from the analysis.
  • Applying our algorithms to a lung adenocarcinoma dataset, hundreds of statistically significant narrow consensus peaks were identified. Many of these peaks reside in the regions of widely-implicated oncogenes and tumour suppressor genes.
  • Our methods detect differential peaks between several phenotypes, most notably between gender, which agrees with known clinic-pathological gender differences in lung cancer. Several positively identified homologues demonstrate the sensitivity of our approach for the detection of copy number change in ultimately narrow, single probe regions.

 

4.   Software tools for diagnostics with re-sequencing technologies for human genome

  •       Re-sequencing technology for human genome has become economically viable within the last year.  NICTA’s partners plan to deploy them in their wet labs within the next few months. In anticipation of this NICTA has developed and applied software tools in pilot projects.  The following two breast cancer datasets collected using Illumina’s Solexa technology were tested:
    • Detection of epigenetic changes in Ductal Carcinoma in situ (DCIS). DCIS accounts for over half of the diagnosed breast lesions. Although they are believed to represent breast cancer precursors, there is currently no successful preventive treatment that is used, beyond anti-estrogens. The cohort will compare those patients that respond with those that didn’t to the standard treatment, by interrogating their epigenetic variations as a complement to pre-existing DNA CNV, mRNA and miRNA profiles.
    • Detection of methylation changes in breast stem cells.  The aim is to develop an improved chemotherapy treatment that targets cancerous stem cells rather than the bulk tumor cells, which have minimal potential for proliferation.

  •       Both experiments have employed pioneering tissue preprocessing techniques.  Initial testing was very positive and further testing is pending.


 

5.   In silico methods for the identification of cis-regulatory modules in humans and its application to cancer

  •  A significant challenge in unraveling the complex machinery behind gene expression is the identification of regulatory elements that control gene expression in the genome. Successful identification of these elements in human cancers can lead to increased understanding of cancer biology and potential therapeutic drug targets.
  •  Traditionally, computational approaches for the identification of regulatory elements in the human genome have focused around the transcription start site (TSS) of genes. Recent high-throughput experimental data demonstrating the genome wide location of some of these elements has shown that less than 20% of these elements reside at the TSS.
  •    NICTA has developed a computational approach that facilitates regulatory element search over large genomic distances which has historically been infeasible. This is achieved through large scale integration of recent high-throughput genomic data (eg. ChIP-SEQ) and the development of novel computational algorithms.
  •   NICTA has integrated our predictions with experimental data provided by Peter MacCallum Cancer Center in Ovarian Cancer and Mouse Tumour models showing the potential for immediate pursuit in the understanding of clinical observations.