| 研究課題/領域番号 |
17K00399
|
|---|
| 研究機関 | お茶の水女子大学 |
|---|
研究代表者 |
オベル加藤 ナタナエル お茶の水女子大学, 基幹研究院, 助教 (10749659)
|
|---|
| 研究期間 (年度) |
2017-04-01 – 2020-03-31
|
| キーワード | DNA Nanostructures / Molecular Robotics / Quality-Diversity / Evolvability / Origins of Life |
|---|
| 研究実績の概要 |
The focus this year has been on the theoretical side, implementing a workflow of increasingly accurate simulations. We used a three-leveled approach: (1) reaction network generation from a set of DNA strands, (2) steady-state structure formation from that reaction network and initial concentrations, and (3) confirmation of structure formation from strands via the oxDNA algorithm.
We implemented a variant of a state-of-the-art Quality-Diversity algorithm named MAP-Elites to explore the space of reaction networks generated by the first simulation level. The algorithm progressively fills a 2D grid, where we store the best set founds for each combination of two features. The first feature represents the average reactivity of the system (total number of reactions divided by the total number of structures produced). The second feature represents the total number of structures produced. Finally, the fitness of a specific set is the size of the largest structure in the reaction network. The algorithm iteratively modifies the systems stored in the grid to generate the best possible solutions for each pair of features, the so-called map of elites. Those features and that fitness were chosen based on theoretical work from the previous year.
The most promising elites are then simulated and initial concentrations optimized to generate a wide variety of structures over staged selection.
Finally, systems that can pass a given fitness threshold undergo the most time-consuming simulation. We have made a list of such candidates for a simple library based on combinatorial 4-domain sequences.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The modeling and algorithmic sides of the project are currently on par with the plan. We now have an algorithmic approach to systematically explore the emergence of complex structures from a specific library of DNA sequences. Moreover, the developed framework can be used in a more general setting than DNA nanostructure emergence and is currently being tested on reaction networks implementing molecular robotics controllers.
On the experimental side, an approach using a repurposed 3D printer to perform the mixing is currently investigated. Pipetting precision is appropriate for the current project, however, the setup does not offer a way to directly observe the system. For that reason, a human step is still required, which prevents full automation. That issue can be mitigated by a stronger focus on the predictive algorithm, minimizing the number of actual in-vitro experiments.
|
| 今後の研究の推進方策 |
As the theoretical side of the project is reaching maturity, the next step is to perform exploration on a much larger scale. We will also investigate simulation and algorithmic parameters to improve the convergence speed of the workflow.
To accelerate further exploration, we plan to add a surrogate model to the workflow. A surrogate model is a predictor trained on available data that helps determine if a given system would have a good fitness value. Since evaluation by a surrogate model is much faster than actual simulation, a properly trained model can greatly improve performance.
On the experimental side, we will continue work on the 3D printer and perform small scale batch of experiments. We will leverage the prediction generated by our model to only focus systems with high predicted fitness. This approach is expected to keep the number of experiment small enough to have a human step in the process. At the same time, we will investigation automation strategies to remove that step.
|
| 次年度使用額が生じた理由 |
Most expenses are going according to plan. A change on the experimental approach that occurred on the first year removed the need to hire a technician to work on the project. We plan to purchase additional computing power with the unused amount to speed up algorithmic exploration. We also plan to purchase electronics parts to improve the automation of the new experimental setup as required.
|
