Research press release


Nature Methods

Visible deep learning reveals how genes shape cell characteristics




Trey Idekerたちは、ディープラーニングアルゴリズムの構造を、細胞内の既知の分子システムの構造にマッピングすることにより、「目に見える」人工ニューラルネットワークを構築した。このモデルは、いったん訓練されると、遺伝的変化の物理的影響を予測できるようになることが明らかになった。さらに、モデルの構成要素が分かっているため、遺伝子と物理的特徴との関係の基盤となる機序について洞察を得ることもできる。研究チームは、こうした目に見えるニューラルネットワークを、遺伝学的論理の理解や、特定の物理的特徴に重要な分子システムの同定、細胞内の新たなプロセスの発見にどのように活用できるかについても示している。

The inner workings of a cell can be revealed by a new deep-learning computer algorithm that has accessible processes, reports a paper published online this week in Nature Methods.

Artificial intelligence can perform many complex tasks typically done by humans, such as recognizing faces, translating languages, and playing games. Deep-learning networks, also known as artificial neural networks, are increasingly being used to automate biological data analysis.

A challenge with deep-learning models is that they are typically ‘black boxes’, meaning that the process by which a model performs a task is not easily identifiable. For biological applications, the ability to examine the way deep-learning models recognize and process the data they analyze could help researchers better understand the biology behind these data.

Trey Ideker and colleagues created a ‘visible’ artificial neural network by mapping the structure of a deep-learning algorithm onto the known structures of molecular systems within cells. The authors show that once the model is trained, it can predict the physical effects of genetic changes. Moreover, as the model’s components are accessible, it can also provide insight about the mechanisms underlying the relationship between genes and physical characteristics. The researchers also show how such a visible neural network can be used to understand genetic logic, identify which molecular systems are important for certain physical characteristics, and discover new processes within the cell.

doi: 10.1038/nmeth.4627


メールマガジンリストの「Nature 関連誌今週のハイライト」にチェックをいれていただきますと、毎週各ジャーナルからの最新の「注目のハイライト」をまとめて皆様にお届けいたします。