Raman Spectroscopy And Machine Learning For Prediction And Characterization Of Stem Cells During Ips Reprogramming

A number of investigations has addressed the use of label-free spontaneous Raman spectroscopy to monitor the differentiation process in stem cells as an alternative to destructive or invasive methods. In this study, using label-free imaging Raman spectroscopy combined with intelligent algorithms, we analysed the reprograming process in an isogenic mouse cell line and compared reprogrammed cells with the original embryonic stem cells (ES) and differentiated cells. Neural network, regression models, and ratiometric analyses were used to discriminate the cell states and extract several important biomarkers specific to differentiation or reprogramming. After appropriate processing, the spectral signatures of single living cells were used as an input of neural network model. Results show a high accuracy in predicting the ES cells, neuronal cells, and cells after 10 or 20 days of reprogramming (with sensitivity and specificity scores of 92% or more). By contrast, an intermediate cell state (5 days of reprogramming) exhibited a lower predictive accuracy (66%) which suggest this population is highly variable in terms of metabolic signatures. Linear regression models and ratiometric analyses confirmed this result and allowed us to extract several spectral wavelengths to characterize and predict the different states. Interestingly, our results make a strong case for substantial differences between stem cells (ES) and cells undergoing reprogramming, notably through the contribution of the lipid band at 1445 cm−1 which is associated to synthesis of fatty acids. This difference was seen even in cells after 20 days of reprogramming which are positive for pluripotency markers (Nanog, Oct4, Sox2 and SSEA-1). Finally, we also showed that the laser does not damage living cells, underlying the technical advantage of our method.