2016 Fiscal Year Annual Research Report
| Project/Area Number |
15F15776
|
|---|
| Research Institution | Institute of Physical and Chemical Research |
|---|
Principal Investigator |
角田 達彦 国立研究開発法人理化学研究所, 統合生命医科学研究センター, グループディレクター (10273468)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
KAMOLA PIOTR 国立研究開発法人理化学研究所, 統合生命医科学研究センター, 外国人特別研究員
|
|---|
| Project Period (FY) |
2015-11-09 – 2018-03-31
|
| Keywords | Personalised Medicine / Multi-Omics / Drug Design / Machine Learning / Network Analysis |
|---|
| Outline of Annual Research Achievements |
During the fellowship we participated in DREAM competitions which aim at answering fundamental questions about translation medicine. For the Drug Combination Prediction Challenge we utilised GBM and LASSO to predict multi-omics, network and signalling biomarkers that predict which chemotherapy drugs work in a synergistic fashion. For the Disease Module Identification Challenge we used a combination of several network topology algorithms, node and edges filtering strategies (based on centrality ranking and connectivity) and Jaccard similarity coefficient to identify clusters of genes that underline complex diseases. The methodology was used with protein-protein, signalling, coexpression, cancer and homology-based networks. For the second sub-challange we also incorporated several methods to combine networks that share information across them. Another project involved tackling small sample size and high dimensionality problem. Specifically, we developed an algorithm that is geared towards such datasets by taking advantage of information based on data distribution, folding the information into feature matrix and modifying expectation-maximization algorithm. The method beats state of the art clustering algorithms for both RNA and methylation datasets. I also collaborated with UMass Medical School, predicting drug-RNA interactions for their tested compounds as well as designing and ranking novel therapeutics for two clinically relevant genes. I am also continuing the collaboration with ETH Zurich on online aptamer database and the CREST project on predictive biomarker discovery.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
Successfully participated in AstraZeneca-Sanger Drug Combination Prediction and Disease Module Identification Challenges. For the later, our team was the top performer in the second sub-challenge (network overlap predictions) and was invited to present a poster and gave a podium talk during RECOMB/ISCB Conference on Regulatory & Systems Genomics (Phoenic, USA, November 2016). Both challenges will lead to publications that will be prepared and submitted by the organisers and will include out team.
The collaboration on clustering algorithm lead to a manuscript which is currently under review in Pattern Recognition journal (with myself as a co-author).
Established and continuing collaboration on drug design and development (UMass Medical School, GlaxoSmithKline, AstraZenneca) and developing online aptamer database (ETH Zurich, Stanford University). The predictions I generated will be used to investigate mouse in vivo drug toxicity and the top-ranking designs will be progressed to pre-clinical testing.
Identified tens of mutations and gene expression patterns within the lung adenocarcinoma dataset that differ between the chemotherapy responder and non-responder groups.
|
| Strategy for Future Research Activity |
Currently we are working on an unsupervised learning and feature extraction method using tensor transformation. The project utilises thousands of in vitro drug screening data and multi-omics datasets for training and validation purposes. The aim is to extract cross-layer feature combination that correlate with IC50s score within a given tissue of interest. The in vitro dataset will be further used to establish novel bioinformatic methods geared toward multi-omics interactions.
I am also progressing the work on multi-omics analysis of patient data. Several methodologies are being explored (supervised, semi-supervised and unsupervised learning, deep learning, statical analysis, novel algorithms) to find genomic features that correlate with clinical covariates, therapeutic response and overall and disease-free survival. We have identified tens of mutation and expression patterns that are currently being tested as predictive biomarkers. Optimal analysis flow is being established to tackle the high heterogeneity and small sample size/very high dimensionality.
The current collaboration on drug design and aptamers will be continued. Furthermore, I will utilise pre-clinical ASO-drug screening data to establish methodology and parameters to predict optimal therapeutic binding site within RNA secondary structure. The approach will be combined with drug selectivity methods that I established in the past to create an end-to-end platform for antisense design and development.
|
Research Products
(4 results)
