한국제약바이오협회

본 DB에서는 해당 논문의 원본데이터 및 이를 활용한 다른 논문의 전처리 데이터의 정보를 제공하고 있습니다.

원저작물에 대한 권리는 해당 연구자 및 기관에 있습니다.

Large-scale drug discovery and repurposing is challenging. Identifying the mechanism of action (MOA) is crucial, yet current approaches are costly and low-throughput. Here we present an approach for MOA identification by profiling changes in mitochondrial phenotypes. By temporally imaging mitochondrial morphology and membrane potential, we established a pipeline for monitoring time-resolved mitochondrial images, resulting in a dataset comprising 570,096 single-cell images of cells exposed to 1,068 United States Food and Drug Administration-approved drugs. A deep learning model named MitoReID, using a re-identification (ReID) framework and an Inflated 3D ResNet backbone, was developed. It achieved 76.32% Rank-1 and 65.92% mean average precision on the testing set and successfully identified the MOAs for six untrained drugs on the basis of mitochondrial phenotype. Furthermore, MitoReID identified cyclooxygenase-2 inhibition as the MOA of the natural compound epicatechin in tea, which was successfully validated in vitro. Our approach thus provides an automated and cost-effective alternative for target identification that could accelerate large-scale drug discovery and repurposing.

Dataset preparation

The mitochondrial phenotypes dataset used in this study encompasses a comprehensive collection of data comprising 1,068 FDA-approved drugs and DMSO. It consists of 35,631 single-cell image sequences, each with 16 timepoints, resulting in a total of 570,096 images derived from three experimental set-ups (see Supplementary Fig. 1 for details on dataset preparation). These drugs have been categorized into 232 classes on the basis of their MOA annotations. To account for the inherent data imbalance across MOA classes, our analysis focuses specifically on 38 classes of MOA (including DMSO) that possess a minimum of five distinct drugs. This selection results in a subset of 16,699 single-cell image sequences, totaling 267,184 images (477 drugs) for further investigation (Supplementary Fig. 6). See Supplementary Data 3 for details on the drug–target relationship and the 38 classes. We split the dataset randomly into training and testing sets at an 8:2 ratio. From the testing set, one sample per MOA category was randomly selected to form the query set, with the remaining samples forming the gallery set.

Annotation of drug–target relationship

Each of the 1,068 compounds included in the commercial library—as well as the eight novel compounds—underwent a rigorous manual annotation process to assign them with specific MOA descriptors. The MOA descriptors were carefully chosen to describe the molecular targets of the compounds, ensuring a higher level of specificity compared with assigning compounds to broader cellular pathways (Supplementary Data 3). Examples of these MOA descriptors include ‘ACE inhibitor’, ‘ Glucocorticoid receptor inhibitors’ and ‘Serotonin receptor antagonist’, among others. Notably, for compounds that exhibited activity against multiple functionally distinct targets, such as kinase inhibitors, multiple MOA descriptors were assigned to accurately represent their complex MOAs. To guide this annotation process, we referred to two past efforts focusing on the manual annotation of FDA-approved drugs^50,51. These annotation efforts provided valuable insights and served as a reference to ensure consistency and accuracy in the assignment of MOA descriptors to the compounds in our study. To avoid the situation of label conflict, we referenced three databases: the Drug Repurposing Hub⁵¹, ChEMBL⁵² and Drugbank⁵³, and prioritized consensus annotations found across all three sources. Note that drugs whose MOA is unrelated to eukaryotic cells, such as those targeting bacteria or viruses, were excluded from annotation.