AI identifies characteristic pattern in tumor cells
2022
Max Delbrück Center for Molecular Medicine in the Helmholtz Association (MDC), Berlin, Germany
Tumors are complex cell tissues that are characterized by high variability, which makes it considerably more difficult to research and identify relevant gene sequences. The present study presents a newly developed algorithm called Icarus, which analyzes differences between cancer cells and the surrounding tissue across different cancer types and data sets. For this purpose, the machine-learning model was fed with numerous data provided to researchers worldwide by various institutions. In the first step, the program created gene signatures and a tumor classifier, through which the algorithm learned to distinguish carcinogenic from healthy cells. Subsequently, the AI was "trained" with various cancer tissue data and the performance of the model was evaluated. Ikarus showed a significantly higher analytical precision than previously developed in silico methods. The model enables not only the characterization of variable, stable cell states but also the functional annotation of individual cells, such as the prediction of differentiation potential, susceptibility to disorders and the prognosis of cell-cell interactions. In all cancer cell types, a characteristic gene sequence pattern has been identified that could provide new insights for cause research. Within this sequence, the algorithm classified certain genes as carcinogenic that have not previously been associated with the development of tumors. The model could prove to be a helpful diagnostic tool and improve therapeutic approaches.
Identifying tumor cells at the single-cell level using machine learning
Altuna Akalin, Verdran Franke
Added on: 08-09-2022
[1] https://genomebiology.biomedcentral.com/articles/10.1186/s13059-022-02683-1[2] https://www.bionity.com/en/news/1176478/ai-identifies-cancer-cells.html
