Аннотация:The outcomes of radiotherapy (RT) of cancer patients significantly depend on the radiosensitivity of tumor to ionizing radiation. The degree of radiosensitivity (RS) and radioresistance (RR) of the tumor is a clinical predictor of the therapeutic response of the tumour in oncopatients to RT and should be considered as a key factor in RT treatment plans in defining the delivered dose, fractionation, and the duration of the RT course. It has been found that cancer cells possess a wide range of radiosensitivity which can vary from high radiosensitivity with high contrast to RS of healthy cells to low radiosensitivity up to radioresistance (RR) leading to poor outcomes of RT. It has been established that the RS of cancer cells depends not only on the ionizing radiation (IR) dose but also on the tumor genotype and phenotype that define cell's ability to adaptation to IR in particular by repairing damaged DNA when exposed to ionizing radiation. Achievement of the durable therapeutic response of patients to RT needs the inclusion molecular and genetic diagnostics in the RT treatment and the development of prognostic and predictive biomarkers of the beneficial RT outcomes. In this direction, the application of machine learning (ML) and artificial intelligence for the prediction of treatment outcomes is a promising way in further development of personalized and precision radiotherapy.
In this work, we developed a method for classification of RS and RR cancer cells based on the analysis of experimental data on clonogenic survival of cancer cells using machine learning. To determine the parameters of cell survival, the experimental dose-effect data were approximated by the linear-quadratic (LQ) model widely used in radiobiology. The dataset included 60 cancer cell lines of different types of cancer such as human pancreatic cancer, colon cancer, lung cancer, breast cancer and others. The dataset included radiosensitive cell lines like Capan-2, DanG (pancreatic cancer), MCF-7, ZR-751 (breast cancer), as well as radioresistance cell lines like suit-2 007, patu-8998T (pancreatic cancer), HPDE (human pancreatic duct epithelial cell line), BT-20 (breast cancer), and others. As a result of experimental data approximation by the LQ model, two parameters 𝛼 and 𝛽 were determined, the ratio of which is commonly used in radiobiology to evaluate the radiosensitivity of cancer cells. A high value of the 𝛼/𝛽 ratio is characteristic of RS cells, which have a low ability to repair damage after IR and vice versa. In order to increase the reliability of discrimination between RS and RR cells according to clonogenic survival data, a combination of the k-means and hierarchal clustering methods along with the principal component analysis was applied. Based on the obtained results, a statistical model was developed and trained on a dataset of experimental data to determine the radiosensitive and radioresistance cancer cells and was successfully validated using the new dataset of parameters α/β and α of cells which were not included in the training dataset.
To investigate association between genetic alterations and radioresistance of the selected cancer cells, we performed bioinformatics analysis of the mutations in cancer cells belonging to the RS and RR clusters. Gene data were derived from the mutation databases COSMIC (sanger.ac), GeneCards (www.genecards.org ) and DepMap (depmap.org). The analysis showed that mutations in the genes code proteins of the key cellular signaling pathways such as 1) repair of DNA damage (TP53, ATM, BRCA1 genes, etc.); 2) cell proliferation (EGFR, PTEN, PI3K, BRAF, etc.), and 3) apoptosis (BCL, BAX, etc.). As established previously the analyzed mutations are responsible for the occurrence of radioresistance in RT patients.
The statistical model for classification of radiosensitive and radioresistant cancer cells was developed and trained on a dataset of experimental data on clonogenic survival of cancer cells under ionizing radiation. The model validation showed that the clustering method satisfactorily classifies cells according to their radiosensitivity. Application of the proposed model to classify the radiosensitivity of cancer cells and determination of radioresistant cell lines can be used to optimize radiotherapy treatment plans in personalized therapy. The further development of the ML model aims at increasing training dataset of cancer cells and extension of the model to the analysis of radiosensitivity of heterogeneous tumors.