Skip to main content

Advanced Data Mining Procedures Applied to Raman Spectroscopy Investigations of Interactions between Drugs and Cells.

Periodic Reporting for period 1 - Spectro-Metrics (Advanced Data Mining Procedures Applied to Raman Spectroscopy Investigations of Interactions between Drugs and Cells.)

Reporting period: 2018-05-28 to 2020-05-27

Spectro-metrics aimed to develop new data mining methodologies for the investigation of drug uptake and cellular responses using Raman Spectroscopy (RS). In short, RS is a very useful technique to i) track a drug within the cell and ii) capture phenotypical changes produced by the drug in the cell in terms of molecular composition. Therefore, this approach has an outstanding potential to study drug kinetics, mechanisms and modes of action, as well as cellular responses and toxicity, thus providing an approach to improve chemotherapy efficiency with promising applications in personalized medicine.
However, the information in the Raman Spectrum is encrypted within a set of overlapped bands from different components, including the drug itself, as well as proteins, lipids, carbohydrates, and DNA from the cell. This makes the extraction of the interpretable data relative to the drug kinetics and cellular responses challenging. The project focused on the investigation of advanced data mining techniques such as multivariate Curve Resolution-Alternate Least Squares (MCR-ALS) to isolate the signal of the drug and the different chemical components of the cell, and thus monitor the drug uptake, chemical binding and subsequent cellular responses, kinetically.
The primary finding of the project was that MCR-ALS was indeed able to identify both the signals of drug, Doxorubicin (DOX, see Figure 2), and the subsequent cellular response. This allowed us to follow the kinetics as well as changes in the DNA and proteins induced by the presence of the drug. In this sense, changes in the DNA were synchronous with the signal of the drug, suggesting intercalation of DOX. In comparison to Partial Least Squares Regression (PLSR), MCR-ALS provided much more information.
Following the success of MCR-ALS, further research was performed to increase the performance of the method. In collaboration with Dr Roma Tauler from the Centre of Scientific Research (CSIC) in Barcelona, we investigated the use of a hard-soft model which uses kinetic equations to guide MCR-ALS iterations. This novel improvement provided a better understanding of the kinetics. Furthermore, it provided a new way to calculate rates for both, drug uptake and subsequent cellular response.
In the second year of the project, and considering that we found some challenges on obtaining information about the biochemical components in the Raman spectra, we explored the use of Infrared (IR) spectroscopy as a complementary technique. A multimodal study was carried out in collaboration with Professor Kamilla Malek at the Jagiellonian University in Krakow, Poland. We evidenced that the combination of both techniques provided a more comprehensive picture of the system, including the drug uptake and intercalation (more visible in RS) and cellular responses (more visible in IR). This has opened up several new research lines, including collaborations with IR research entities such as the Company Photothermal (United States of America) and the Soleil Synchrotron (France).
The research carried out was disseminated in peer-review publications, some manuscripts being still in preparation. Other outreach activities included a conference talk, being invited as a speaker in a Workshop in Poland and tutoring one master and one PhD student. In addition, I am editing a special issue in an international Peer Review journal aiming to promote statistical methods to treat spectral data with clinical applications. All publications within the context of the project have been stored at the public repository of the beneficiary, ( and datasets have been made openly available in the Zenodo repository, thus compiling with Horizon 2020 guidelines.
Results obtained during Spectro-metrics have yield new methodologies for the study of drug mode of action and cellular responses using vibrational spectroscopy. Before the project, the tools available for treating the data in this context were generally limited. They included, for example, the use of unsupervised data (e.g. PCA), which could fail to extract important information when the spectral features of interest are not the main source of spectral variation. Another example is the use of the regression vector of a PLSR, which, as we demonstrated during the project, could fail to model non-linear responses of data. Spectro-metrics has provided a toolbox of novel methodologies for elucidating drug uptake and cellular responses from Raman and IR data of time dependent incubation experiments, highlighting:
- The validation of MCR-ALS as a technique to extract the concentration profiles and pure spectra of the drug and the biochemical changes induced, including initial binding and subsequent cellular responses.
- Creation of hard-soft models to incorporate within the process the kinetic equations, improving the interpretability of the model and providing an estimation of the equation constants.
- Methods for the treatment of multimodal vibrational data, allowing to combine the signals from Raman (More sensitive to resonant drugs) and IR (More sensitive to changes in the biochemical composition of the cell). Figure 3 depicts the cross correlation map for Raman and IR spectra for a incubation time experiment.
- The development of “ATR-Spin” concept to be able to measure the IR spectra of the cells with a single step of preconcentration and isolation, thus enabling the measurement without the need of an expensive FTIR microscope.

In general, the set of methodologies created is going to expand our knowledge about cellular processes. For example, works are in due course to study the beta-oxydation process in hepatic cells. In the case of the study of drugs, the better understanding of modes and mechanisms of action will assist in the development of better and personalized therapies against cancer.
Doxorubicin Structure
Cross Correation
A459 cell under the microscope
Equations Used to model the drug uptake