Phytochemicals in Pancreatic Cancer Treatment: A Machine Learning Study

The discovery of new strategies and novel therapeutic agents is crucial to improving the current treatment methods and increasing the efficacy of cancer therapy. Phytochemicals, naturally occurring bioactive constituents derived from plants, have great potential in preventing and treating various diseases, including cancer. This study reviewed 74 literature studies published between 2006 and 2022 that conducted in vitro cytotoxicity and cell apoptosis analyses of the different concentrations of phytochemicals and their combinations with conventional drugs or supplementary phytochemicals on human pancreatic cell lines. From 34 plant-derived phytochemicals on 20 human pancreatic cancer cell lines, a total of 11 input and 2 output variables have been used to construct the data set that contained 2161 different instances. The machine learning approach has been implemented using random forest for regression, whereas association rule mining has been used to determine the effects of individual phytochemicals. The random forest models developed are generally good, indicating that the phytochemical type, its concentration, and the type of cell line are the most important descriptors for predicting the cell viability. However, for predicting cell apoptosis the primary phytochemical type is the most significant descriptor . Among the studied phytochemicals, catechin and indole-3-carbinol were found to be non-cytotoxic at all concentrations irrespective of the treatment time. On the other hand, berbamine and resveratrol were strongly cytotoxic with cell viabilities of less than 40% at a concentration range between 10 and 100 μM and above 100 μM, respectively, which brings them forward as potential therapeutic agents in the treatment of pancreatic cancer.

High cytotoxicity and low viability of cancer cells are necessary for effective treatment solutions.Hence, for each phytochemical type, the average viability percentage was calculated using various concentrations of the specific phytochemical irrespective of the cell line type.
After applying this procedure to each of the 34 phytochemicals, the overall average cell viability percentage is found as 63.7%.The average cell viability percentages for the most studied six phytochemicals, i.e., baicalein, curcumin, escin, resveratrol, xanthohumol, and mangostin, are given in Figure S3.Among these, baicalein, curcumin, escin, and resveratrol resulted in lower viability percentages than the overall average cell viability, with values of 53.4%, 51.4%, 52.9%, and 56.8%, respectively.-mangostin has an average viability percentage of 63.6% which is close to the overall average.The average viability percentage of the cells exposed to xanthohumol is 67.4% which is relatively higher than the overall average.
Consequently, lower viability percentages obtained with baicalein, curcumin, escin, and resveratrol present promising results for potential treatment options in human pancreatic cancer.For all phytochemicals with average viability results that are lower than the overall average viability are given in Figure S4.The most frequently studied phytochemicals are α-mangostin, baicalein, curcumin, escin, resveratrol, and xanthohumol.When their pathways are analyzed in detail, it is seen that α-mangostin can inhibit many pathways, and the distribution of the data numbers are similar.Baicalein mainly activates caspase-3/PARP pathway but also inhibits MEK/ERK and PI3K/Akt pathways.Curcumin also mainly activates caspase-3/PARP pathway, but some studies reported its contribution to the inhibition of ERK 1/2 as well (this happened only when Sulforaphane and aspirin were used as supplementary phytochemicals and drugs, respectively).Escin inhibits NF-kB activity only.Resveratrol was reported to activate Nrf2 and inhibit ERK and sonic hedgehog signaling pathways.Lastly, Xanthohumol mainly causes inhibition of STAT3, which is followed by activation of Nrf2 and inhibition of Notch1.
Medium, Viability Assay, Viability and Apoptosis DistributionsThe cell culture medium is analyzed for 20 human pancreatic cell lines, and five culture media are recorded.The growth mediums including Dulbecco's Modified Eagle's Medium (DMEM), Roswell Park Memorial Institute 1640 Medium (RPMI-1640), Leibovitz's (L-15) medium, McCoy's 5A medium, and Iscove's Modified Dulbecco's Medium (IMDM) are present in our dataset.Figure S1 shows the count of data points for DMEM, L-15, and RPMI-1640 mediums for the cell lines used.IMDM was solely used for CFPAC-1 cell line, and similarly, McCoy's 5A medium was used only for CAPAN-2 cell line.Thus, their graphs were not shown in Figure S1.DMEM is the most preferred cell culture medium for PANC-1 cell line, and RPMI-1640 is the most preferred for BxPC-3 cell line.Different cell viability assay types can measure cell viability.FigureS2shows the cell viability assay types distribution among the investigated cases in this study.dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide) assay is the most prominent assay type.With 2071 data points, it constitutes 52% of the total cases.The second most used assay type within this study is CCK-8 (Cell Counting Kit-8) which includes 17% of the cases.This is followed by MTS (3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4sulfophenyl)-2H-tetrazolium) assay, which constitutes 15% of the cases.

Figure S1 .
Figure S1.The distribution of a)cell culture medium among all cells, and the distribution of b) DMEM c) L-15 Medium, d) RPMI-1640 cell culture mediums among different pancreatic cell lines.

Figure S2 .
Figure S2.Distribution of cell viability assay type.

Figure S4 .
Figure S4.Average cell viability percentages versus phytochemical type graph with lower viability percentages compared to the average value.

Figure S5 .
Figure S5.Phytochemical vs Average Cell Apoptosis (%) based on the number of data points.

Figure S7 .
Figure S7.Cell Signaling pathways related to main phytochemicals reported in the dataset.