Group and Basis Restricted Non-Negative Matrix Factorization and Random Forest for Molecular Histotype Classification and Raman Biomarker Monitoring in Breast Cancer

Stratification of tumour cell radiation response and metabolic signatures visualization with Raman spectroscopy and explainable convolutional neural network

Reprogramming of cellular metabolism is a driving factor of tumour progression and radiation therapy resistance. Identifying biochemical signatures associated with tumour radioresistance may assist with the development of targeted treatment strategies to improve clinical outcomes. Raman spectroscopy (RS) can monitor post-irradiation biomolecular changes and signatures of radiation response in tumour cells in a label-free manner. Convolutional Neural Networks (CNN) perform feature extraction directly from data in an end-to-end learning manner, with high classification performance. Furthermore, recently developed CNN explainability techniques help visualize the critical discriminative features captured by the model. In this work, a CNN is developed to characterize tumour response to radiotherapy based on its degree of radioresistance. The model was trained to classify Raman spectra of three human tumour cell lines as radiosensitive (LNCaP) or radioresistant (MCF7, H460) over a range of treatment doses and data collection time points. Additionally, a method based on Gradient-Weighted Class Activation Mapping (Grad-CAM) was used to determine response-specific salient Raman peaks influencing the CNN predictions. The CNN effectively classified the cell spectra, with accuracy, sensitivity, specificity, and F1 score exceeding 99.8%. Grad-CAM heatmaps of H460 and MCF7 cell spectra (radioresistant) exhibited high contributions from Raman bands tentatively assigned to glycogen, amino acids, and nucleic acids. Conversely, heatmaps of LNCaP cells (radiosensitive) revealed activations at lipid and phospholipid bands. Finally, Grad-CAM variable importance scores were derived for glycogen, asparagine, and phosphatidylcholine, and we show that their trends over cell line, dose, and acquisition time agreed with previously established models. Thus, the CNN can accurately detect biomolecular differences in the Raman spectra of tumour cells of varying radiosensitivity without requiring manual feature extraction. Finally, Grad-CAM may help identify metabolic signatures associated with the observed categories, offering the potential for automated clinical tumour radiation response characterization.

Reconstruction of Raman Spectra of Biochemical Mixtures Using Group and Basis Restricted Non-Negative Matrix Factorization

Raman spectroscopy is a useful tool for obtaining biochemical information from biological samples. However, interpretation of Raman spectroscopy data in order to draw meaningful conclusions related to the biochemical make up of cells and tissues is often difficult and could be misleading if care is not taken in the deconstruction of the spectral data. Our group has previously demonstrated the implementation of a group- and basis-restricted non-negative matrix factorization (GBR-NMF) framework as an alternative to more widely used dimensionality reduction techniques such as principal component analysis (PCA) for the deconstruction of Raman spectroscopy data as related to radiation response monitoring in both cellular and tissue data. While this method provides better biological interpretability of the Raman spectroscopy data, there are some important factors which must be considered in order to provide the most robust GBR-NMF model. We here evaluate and compare the accuracy of a GBR-NMF model in the reconstruction of three mixture solutions of known concentrations. The factors assessed include the effect of solid versus solutions bases spectra, the number of unconstrained components used in the model, the tolerance of different signal to noise thresholds, and how different groups of biochemicals compare to each other. The robustness of the model was assessed by how well the relative concentration of each individual biochemical in the solution mixture is reflected in the GBR-NMF scores obtained. We also evaluated how well the model can reconstruct original data, both with and without the inclusion of an unconstrained component. Overall, we found that solid bases spectra were generally comparable to solution bases spectra in the GBR-NMF model for all groups of biochemicals. The model was found to be relatively tolerant of high levels of noise in the mixture solutions using solid bases spectra. Additionally, the inclusion of an unconstrained component did not have a significant effect on the deconstruction, on the condition that all biochemicals in the mixture were included as bases chemicals in the model. We also report that some groups of biochemicals achieve a more accurate deconstruction using GBR-NMF than others, likely due to similarity in the individual bases spectra.

Raman spectroscopy and convolutional neural networks for monitoring biochemical radiation response in breast tumour xenografts

Tumour cells exhibit altered metabolic pathways that lead to radiation resistance and disease progression. Raman spectroscopy (RS) is a label-free optical modality that can monitor post-irradiation biomolecular signatures in tumour cells and tissues. Convolutional Neural Networks (CNN) perform automated feature extraction directly from data, with classification accuracy exceeding that of traditional machine learning, in cases where data is abundant and feature extraction is challenging. We are interested in developing a CNN-based predictive model to characterize clinical tumour response to radiation therapy based on their degree of radiosensitivity or radioresistance. In this work, a CNN architecture is built for identifying post-irradiation spectral changes in Raman spectra of tumour tissue. The model was trained to classify irradiated versus non-irradiated tissue using Raman spectra of breast tumour xenografts. The CNN effectively classified the tissue spectra, with accuracies exceeding 92.1% for data collected 3 days post-irradiation, and 85.0% at day 1 post-irradiation. Furthermore, the CNN was evaluated using a leave-one-out- (mouse, section or Raman map) validation approach to investigate its generalization to new test subjects. The CNN retained good predictive accuracy (average accuracies 83.7%, 91.4%, and 92.7%, respectively) when little to no information for a specific subject was given during training. Finally, the classification performance of the CNN was compared to that of a previously developed model based on group and basis restricted non-negative matrix factorization and random forest (GBR-NMF-RF) classification. We found that CNN yielded higher classification accuracy, sensitivity, and specificity in mice assessed 3 days post-irradiation, as compared with the GBR-NMF-RF approach. Overall, the CNN can detect biochemical spectral changes in tumour tissue at an early time point following irradiation, without the need for previous manual feature extraction. This study lays the foundation for developing a predictive framework for patient radiation response monitoring.

A Machine Learning Framework for Detecting COVID-19 Infection Using Surface-Enhanced Raman Scattering

In this study, we explored machine learning approaches for predictive diagnosis using surface-enhanced Raman scattering (SERS), applied to the detection of COVID-19 infection in biological samples. To do this, we utilized SERS data collected from 20 patients at the University of Maryland Baltimore School of Medicine. As a preprocessing step, the positive-negative labels are obtained using Polymerase Chain Reaction (PCR) testing. First, we compared the performance of linear and nonlinear dimensionality techniques for projecting the high-dimensional Raman spectra to a low-dimensional space where a smaller number of variables defines each sample. The appropriate number of reduced features used was obtained by comparing the mean accuracy from a 10-fold cross-validation. Finally, we employed Gaussian process (GP) classification, a probabilistic machine learning approach, to correctly predict the occurrence of a negative or positive sample as a function of the low-dimensional space variables. As opposed to providing rigid class labels, the GP classifier provides a probability (ranging from zero to one) that a given sample is positive or negative. In practice, the proposed framework can be used to provide high-throughput rapid testing, and a follow-up PCR can be used for confirmation in cases where the model's uncertainty is unacceptably high.

Nanotube abundance from non-negative matrix factorization of Raman spectra as an example of chemical purity from open source machine learning

The chemical purity of materials is important for semiconductors, including the carbon nanotube material system, which is emerging in semiconductor applications. One approach to get statistically meaningful abundances and/or concentrations is to measure a large number of small samples. Automated multivariate classification algorithms can be used to draw conclusions from such large data sets. Here, we use spatially-mapped Raman spectra of mixtures of chirality-sorted single walled carbon nanotubes dispersed sparsely on flat silicon/silicon oxide substrates. We use non-negative matrix factorization (NMF) decomposition in scikit-learn, an open-source, python language “machine learning” package, to extract spectral components and derive weighting factors. We extract the abundance of minority species (7,5) nanotubes in mixtures by testing both synthetic data, and real samples prepared by dilution. We show how noise limits the purity level that can be evaluated. We determine real situations where this approach works well, and identify situations where it fails.

Diagnosing molecular subtypes of breast cancer by means of Raman spectroscopy

OBJECTIVES

Raman spectroscopy has been used to discriminate human breast cancer and its different tumor molecular subtypes (luminal A, luminal B, HER2, and triple-negative) from normal tissue in surgical specimens.

MATERIALS AND METHODS

Breast cancer and normal tissue samples from 31 patients were obtained by surgical resection and submitted for histopathology. Before anatomopathological processing, the samples had been submitted to Raman spectroscopy (830 nm, 25 mW excitation laser parameters). In total, 424 Raman spectra were obtained. Principal component analysis (PCA) was used in an exploratory analysis to unveil the compositional differences between the tumors and normal tissues. Discriminant models were developed to distinguish the different cancer subtypes by means of partial least squares (PLS) regression.

RESULTS

PCA vectors showed spectral features referred to the biochemical constitution of breast tissues, such as lipids, proteins, amino acids, and carotenoids, where lipids were decreased and proteins were increased in breast tumors. Despite the small spectral differences between the different subtypes of tumor and normal tissues, the discriminant model based on PLS was able to discriminate the spectra of the breast tumors from normal tissues with an accuracy of 97.3%, between luminal and nonluminal subtypes with an accuracy of 89.9%, between nontriple-negative and triple-negative with an accuracy of 94.7%, and each molecular subtype with an accuracy of 73.0%.

CONCLUSION

PCA could reveal the compositional difference between tumors and normal tissues, and PLS could discriminate the Raman spectra of breast tissues regarding the molecular subtypes of cancer, being a useful tool for cancer diagnosis.