Model Parameter identification using 2D vs 3D experimental data: a comparative analysis

Abstract

Computational models are becoming an increasingly valuable tool in biomedical research. They enable the quantification of variables difficult to measure experimentally, an increase in the spatio-temporal resolution of the experiments and the testing of hypotheses. Parameter estimation from in-vitro data, remains a challenge, due to the limited availability of experimental datasets acquired in directly comparable conditions. While the use of computational models to supplement laboratory results contributes to this issue, a more extensive analysis of the effect of incomplete or inaccurate data on the parameter optimization process and its results is warranted. To this end, we compared the results obtained from the same in-silico model of ovarian cancer cell growth and metastasis, calibrated with datasets acquired from two different experimental settings: a traditional 2D monolayer, and 3D cell culture models. The differential behaviour of these models will inform the role and importance of experimental data in the calibration of computational models' calibration. This work will also provide a set of general guidelines for the comparative testing and selection of experimental models and protocols to be used for parameter optimization in computational models.

Accurate identification of cancer cells in complex pre-clinical models using deep-learning: a transfection-free approach

“…The annotations for the training and test datasets were obtained using the procedural segmentation previously described (Cortesi et al, 2023), which relies on a global thresholding using the Otsu's method. The bounding boxes were then determined as the minimum and maximum values of the coordinates of each segmented cell.…”

“…Manual screening of the images was conducted to exclude any incorrectly labelled ones and thus ensure the accuracy of the gold standard. As previously described (Cortesi et al, 2023) the segmented cells were classified as cancer if their area was between 50 and 5,000 pixels and they had an average fluorescence intensity higher than the background.…”

“…Transwell invasion assays were conducted as previously described (Cortesi et al, 2023). Briefly, for monoculture models (dataset ID 2, 5 and training and test sets) Matrigel-precoated Transwell inserts (Corning Life Sciences, US) were seeded with either HGSOC cells (Table 2) or a mixture of fibroblasts and mesothelial cells (Table 2).…”

“…Each region of the image recognised as a cell is outlined by a bounding box, colour-coded according to the corresponding classification, red for cancer and blue for healthy cells. Figure 3b shows the comparison between ORACLE's results and the gold standard values obtained using the procedural segmentation described in (Cortesi et al, 2023). This algorithm relies on morphological and intensity features to distinguish between cancer and healthy cells.…”

“…This model, initially described by Kenny and colleagues (Kenny et al, 2015) combines three different cell types (fibroblast, mesothelial and cancer cells) and mimics the structure of the omentum, the most common dissemination site for ovarian cancer. The three cell types are seeded in standard Transwell inserts with a surrogate extracellular matrix (Figure 1), enabling the quantification of cell invasion (Cortesi et al, 2023;Joshi et al, 2021;Peters et al, 2015b). Cancer cells used in these experiments, however, need to be transfected with a fluorescent protein to avoid overestimating invasion by including migrated healthy cells.…”