2018 Fiscal Year Research-status Report
A Deep Learning framework for cancer precision medicine
| Project/Area Number |
18K18156
|
|---|
| Research Institution | Institute of Physical and Chemical Research |
|---|
Principal Investigator |
Lysenko Artem 国立研究開発法人理化学研究所, 生命医科学研究センター, 特別研究員 (80753805)
|
|---|
| Project Period (FY) |
2018-04-01 – 2020-03-31
|
| Keywords | Deep Learning / cancer / meta-learning / multiomics |
|---|
| Outline of Annual Research Achievements |
The work this year primarily focused on achieving three core tasks (1) purchase and installation of necessary equipment and software (2) development of a cancer immunology use-case (3) design and prototyping of deep learning models suitable for this type of analysis.
As was originally planned, 20-CPU server with Tesla P100 GPU was purchased and configured to run CUDA, TensorFlow and Keras DNN software. To enable comprehensive exploration of HCC multi-omics data from the cancer immunology perspective, secondary analysis was done on the TCGA and data from our collaborators. This included pathway activity-based approaches (e.g. PARADIGM, SPIA); immune cell abundances (e.g. CIBERSORT, EPIC) and immune-related gene signature analysis (e.g. those from the Thorsson et al. (2018) study, 'immunoduct' analysis pipeline and literature). It was also decided that a re-analysis of original mutation calls with the GATK4 pipeline is necessary, which has caused a slight delay. However, it was still possible to do planned DNN survival model prototyping on the publicly-available TCGA consortium data. In particular, techniques to model disease-free and overall survival with DNNs were explored as well as models for predicting particular types of anti-cancer immune response. This work has been successful in identification of high-performance DNN architectures for these tasks and it is expected that these types of models can be used in combination with a meta-learning framework, which will be investigated in the current year as was originally planned.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The project is progressing on schedule as planned, slight delay was caused because it took longer to complete the data preparation than originally planned.
|
| Strategy for Future Research Activity |
Currently the work is being done towards delivering the ultimate objectives of the project (deep meta-learning methodology for cancer multiomics) and preparation of publications detailing the results from the initial analysis already done. Specifically, the main efforts are devoted to the development of an optimal DNN architecture suitable for this task and evaluation of the latest advances/ideas in the field in terms of their potential to improve the performance of the system.
Once this part of the work is complete, the final objective to develop a meta-learning framework will be tackled. Then, as originally planned., four different strategies for achieving best possible performance in the context of cancer multi-omics data will be explored.
Finally, in combination the analysis and developed modelling techniques will be used to realize the practical deliverables of this project. Specifically, the creation of accurate and reliable models for recurrence and survival risk in hepatocellular carcinoma. This aim will be achieved by working with our clinical collaborators to identify and evaluate models and/or combinations of markers most useful for guiding clinical decisions and supporting precision medicine.
|
| Causes of Carryover |
The under-spent money from the first year incurred because there was a very large amount of work to do to complete the planned objectives in the first year, which meant that there was not as much time left for the dissemination of results that was planned originally. Consequently, less was spent on conference fees/travel etc. and publication fees. The outstanding money can therefore be more efficiently spent this year to perform an equipment upgrade (as newer technological solutions have now became available) to speed up the work and to cover the cost of publishing / presenting this work (as more results are generated towards the end of the project).
|
