Deep autoencoder as an interpretable tool for Raman spectroscopy investigation of chemical and extracellular vesicle mixtures.

Surface-enhanced Raman spectroscopy (SERS) is a powerful tool that provides valuable insight into the molecular contents of chemical and biological samples. However, interpreting Raman spectra from complex or dynamic datasets remains challenging, particularly for highly heterogeneous biological samples like extracellular vesicles (EVs). To overcome this, we developed a tunable and interpretable deep autoencoder for the analysis of several challenging Raman spectroscopy applications, including synthetic datasets, chemical mixtures, a chemical milling reaction, and mixtures of EVs. We compared the results with classical methods (PCA and UMAP) to demonstrate the superior performance of the proposed technique. Our method can handle small datasets, provide a high degree of generalization such that it can fill unknown gaps within spectral datasets, and even quantify relative ratios of cell line-derived EVs to fetal bovine serum-derived EVs within mixtures. This simple yet robust approach will greatly improve the analysis capabilities for many other Raman spectroscopy applications.

Manifold Learning Enables Interpretable Analysis of Raman Spectra from Extracellular Vesicle and Other Mixtures.

Extracellular vesicles (EVs) have emerged as promising diagnostic and therapeutic candidates in many biomedical applications. However, EV research continues to rely heavily on in vitro cell cultures for EV production, where the exogenous EVs present in fetal bovine (FBS) or other required serum supplementation can be difficult to remove entirely. Despite this and other potential applications involving EV mixtures, there are currently no rapid, robust, inexpensive, and label-free methods for determining the relative concentrations of different EV subpopulations within a sample. In this study, we demonstrate that surface-enhanced Raman spectroscopy (SERS) can biochemically fingerprint fetal bovine serum-derived and bioreactor-produced EVs, and after applying a novel manifold learning technique to the acquired spectra, enables the quantitative detection of the relative amounts of different EV populations within an unknown sample. We first developed this method using known ratios of Rhodamine B to Rhodamine 6G, then using known ratios of FBS EVs to breast cancer EVs from a bioreactor culture. In addition to quantifying EV mixtures, the proposed deep learning architecture provides some knowledge discovery capabilities which we demonstrate by applying it to dynamic Raman spectra of a chemical milling process. This label-free characterization and analytical approach should translate well to other EV SERS applications, such as monitoring the integrity of semipermeable membranes within EV bioreactors, ensuring the quality or potency of diagnostic or therapeutic EVs, determining relative amounts of EVs produced in complex co-culture systems, as well as many Raman spectroscopy applications.

Cascaded Deep Convolutional Neural Networks as Improved Methods of Preprocessing Raman Spectroscopy Data.

Machine learning has had a significant impact on the value of spectroscopic characterization tools, particularly in biomedical applications, due to its ability to detect latent patterns within complex spectral data. However, it often requires extensive data preprocessing, including baseline correction and denoising, which can lead to an unintentional bias during classification. To address this, we developed two deep learning methods capable of fully preprocessing raw Raman spectroscopy data without any human input. First, cascaded deep convolutional neural networks (CNN) based on either ResNet or U-Net architectures were trained on randomly generated spectra with augmented defects. Then, they were tested using simulated Raman spectra, surface-enhanced Raman spectroscopy (SERS) imaging of chemical species, low resolution Raman spectra of human bladder cancer tissue, and finally, classification of SERS spectra from human placental extracellular vesicles (EVs). Both approaches resulted in faster training and complete spectral preprocessing in a single step, with more speed, defect tolerance, and classification accuracy compared to conventional methods. These findings indicate that cascaded CNN preprocessing is ideal for biomedical Raman spectroscopy applications in which large numbers of heterogeneous spectra with diverse defects need to be automatically, rapidly, and reproducibly preprocessed.

Classification of Preeclamptic Placental Extracellular Vesicles Using Femtosecond Laser Fabricated Nanoplasmonic Sensors.

Placental extracellular vesicles (EVs) play an essential role in pregnancy by protecting and transporting diverse biomolecules that aid in fetomaternal communication. However, in preeclampsia, they have also been implicated in contributing to disease progression. Despite their potential clinical value, current technologies cannot provide a rapid and effective means of differentiating between healthy and diseased placental EVs. To address this, a fabrication process called laser-induced nanostructuring of SERS-active thin films (LINST) was developed to produce scalable nanoplasmonic substrates that provide exceptional Raman signal enhancement and allow the biochemical fingerprinting of EVs. After validating the performance of LINST substrates with chemical standards, placental EVs from tissue explant cultures were characterized, demonstrating that preeclamptic and normotensive placental EVs have classifiably distinct Raman spectra following the application of advanced machine learning algorithms. Given the abundance of placental EVs in maternal circulation, these findings encourage immediate exploration of surface-enhanced Raman spectroscopy (SERS) of EVs as a promising method for preeclampsia liquid biopsies, while this novel fabrication process will provide a versatile and scalable substrate for many other SERS applications.

A versatile cancer cell trapping and 1D migration assay in a microfluidic device.

Highly migratory cancer cells often lead to metastasis and recurrence and are responsible for the high mortality rates in many cancers despite aggressive treatment. Recently, the migratory behavior of patient-derived glioblastoma multiforme cells on microtracks has shown potential in predicting the likelihood of recurrence, while at the same time, antimetastasis drugs have been developed which require simple yet relevant high-throughput screening systems. However, robust in vitro platforms which can reliably seed single cells and measure their migration while mimicking the physiological tumor microenvironment have not been demonstrated. In this study, we demonstrate a microfluidic device which hydrodynamically seeds single cancer cells onto stamped or femtosecond laser ablated polystyrene microtracks, promoting 1D migratory behavior due to the cells' tendency to follow topographical cues. Using time-lapse microscopy, we found that single U87 glioblastoma multiforme cells migrated more slowly on laser ablated microtracks compared to stamped microtracks of equal width and spacing (p < 0.05) and exhibited greater directional persistence on both 1D patterns compared to flat polystyrene (p < 0.05). Single-cell morphologies also differed significantly between flat and 1D patterns, with cells on 1D substrates exhibiting higher aspect ratios and less circularity (p < 0.05). This microfluidic platform could lead to automated quantification of single-cell migratory behavior due to the high predictability of hydrodynamic seeding and guided 1D migration, an important step to realizing the potential of microfluidic migration assays for drug screening and individualized medicine.

Photonic structures in diamond based on femtosecond UV laser induced periodic surface structuring (LIPSS).

We study the fabrication of photonic surface structures in single crystal diamond by means of highly controllable direct femtosecond UV laser induced periodic surface structuring. By appropriately selecting the excitation wavelength, intensity, number of impinging pulses and their polarization state, we demonstrate emerging high quality and fidelity diamond grating structures with surface roughness below 1.4 nm. We characterize their optical properties and study their potential for the fabrication of photonic structure anti-reflection coatings for diamond Raman lasers in the near-IR.