Lluís Fernández RomeroPostdoctoral Researcher
Signal and Information Processing for Sensing Systems
+34 934 031 118
Staff member publications
Mas, S., Torro, A., Bec, N., Fernández, L., Erschov, G., Gongora, C., Larroque, C., Martineau, P., de Juan, A., Marco, S., (2019). Use of physiological information based on grayscale images to improve mass spectrometry imaging data analysis from biological tissues Analytica Chimica Acta In Press, Accepted Manuscript
The characterization of cancer tissues by matrix-assisted laser desorption ionization-mass spectrometry images (MALDI-MSI) is of great interest because of the power of MALDI-MS to understand the composition of biological samples and the imaging side that allows for setting spatial boundaries among tissues of different nature based on their compositional differences. In tissue-based cancer research, information on the spatial location of necrotic/tumoral cell populations can be approximately known from grayscale images of the scanned tissue slices. This study proposes as a major novelty the introduction of this physiologically-based information to help in the performance of unmixing methods, oriented to extract the MS signatures and distribution maps of the different tissues present in biological samples. Specifically, the information gathered from grayscale images will be used as a local rank constraint in Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) for the analysis of MALDI-MSI of cancer tissues. The use of this constraint, setting absence of certain kind of tissues only in clear zones of the image, will help to improve the performance of MCR-ALS and to provide a more reliable definition of the chemical MS fingerprint and location of the tissues of interest. The general strategy to address the analysis of MALDI-MSI of cancer tissues will involve the study of the MCR-ALS results and the posterior use of MCR-ALS scores as dimensionality reduction for image segmentation based on K-means clustering. The resolution method will provide the MS signatures and their distribution maps for each tissue in the sample. Then, the resolved distribution maps for each biological component (MCR scores) will be submitted as initial information to K-means clustering for image segmentation to obtain information on the boundaries of the different tissular regions in the samples studied. MCR-ALS prior to K-means not only provides the desired dimensionality reduction, but additionally resolved non-biological signal contributions are not used and the weight given to the different biological components in the segmentation process can be modulated by suitable preprocessing methods.
Keywords: MCR-ALS, K-means, Local rank constraints, MALDI-MSI, Grayscale images
Contreras, M. D. M., Jurado-Campos, N., Sánchez-Carnerero Callado, C., Arroyo-Manzanares, N., Fernández, L., Casano, S., Marco, S., Arce, L., Ferreiro-Vera, C., (2018). Thermal desorption-ion mobility spectrometry: A rapid sensor for the detection of cannabinoids and discrimination of Cannabis sativa L. chemotypes Sensors and Actuators B: Chemical 273, 1413-1424
Existing analytical techniques used for the determination of cannabinoids in Cannabis sativa L. (Cannabis) plants mostly rely on chromatography-based methods. As a rapid alternative for the direct analysis of them, thermal desorption (TD)-ion mobility spectrometry (IMS) was used for obtaining spectral fingerprints of single cannabinoids from Cannabis plant extracts and from plant residues on hands after their manipulation. The ionization source was 63Ni, with automatic switchable polarity. Although in both ionization modes there were signals in the TD-IMS spectra of the plant extracts and residues that could be assigned to concrete cannabinoids and chemotypes, most of them could not be clearly distinguished. Alternatively, the global spectral data of the plant extracts and residues were pre-processed and then, using principal component analysis (PCA)-linear discriminant analysis (LDA), grouped in function of their chemotype in a more feasible way. Using this approach, the possibility of false positive responses was also studied analyzing other non-Cannabis plants and tobacco, which were clustered in a different group to those of Cannabis. Therefore, TD-IMS, as analytical tool, and PCA-LDA, as a strategy for data reduction and pattern recognition, can be applied for on-site chemotaxonomic discrimination of Cannabis varieties and detection of illegal marijuana since the IMS equipment is portable and the analysis time is highly short.
Keywords: Cannabis sativa L., Cannabinoids, Chemometrics, ChemotypeIon mobility spectrometry
Rodríguez-Pérez, R., Fernández, L., Marco, S., (2018). Overoptimism in cross-validation when using partial least squares-discriminant analysis for omics data: a systematic study Analytical and Bioanalytical Chemistry 410, (23), 5981-5992
Advances in analytical instrumentation have provided the possibility of examining thousands of genes, peptides, or metabolites in parallel. However, the cost and time-consuming data acquisition process causes a generalized lack of samples. From a data analysis perspective, omics data are characterized by high dimensionality and small sample counts. In many scenarios, the analytical aim is to differentiate between two different conditions or classes combining an analytical method plus a tailored qualitative predictive model using available examples collected in a dataset. For this purpose, partial least squares-discriminant analysis (PLS-DA) is frequently employed in omics research. Recently, there has been growing concern about the uncritical use of this method, since it is prone to overfitting and may aggravate problems of false discoveries. In many applications involving a small number of subjects or samples, predictive model performance estimation is only based on cross-validation (CV) results with a strong preference for reporting results using leave one out (LOO). The combination of PLS-DA for high dimensionality data and small sample conditions, together with a weak validation methodology is a recipe for unreliable estimations of model performance. In this work, we present a systematic study about the impact of the dataset size, the dimensionality, and the CV technique used on PLS-DA overoptimism when performance estimation is done in cross-validation. Firstly, by using synthetic data generated from a same probability distribution and with assigned random binary labels, we have obtained a dataset where the true classification rate (CR) is 50%. As expected, our results confirm that internal validation provides overoptimistic estimations of the classification accuracy (i.e., overfitting). We have characterized the CR estimator in terms of bias and variance depending on the internal CV technique used and sample to dimensionality ratio. In small sample conditions, due to the large bias and variance of the estimator, the occurrence of extremely good CRs is common. We have found that overfitting peaks when the sample size in the training subset approaches the feature vector dimensionality minus one. In these conditions, the models are neither under- or overdetermined with a unique solution. This effect is particularly intense for LOO and peaks higher in small sample conditions. Overoptimism is decreased beyond this point where the abundance of noisy produces a regularization effect leading to less complex models. In terms of overfitting, our study ranks CV methods as follows: Bootstrap produces the most accurate estimator of the CR, followed by bootstrapped Latin partitions, random subsampling, K-Fold, and finally, the very popular LOO provides the worst results. Simulation results are further confirmed in real datasets from mass spectrometry and microarrays.
Keywords: Metabolomics, Mass spectrometry, Microarrays, Chemometrics, Data analysis, Classification, Method validation
Fonollosa, J., Fernández, L., Gutiérrez-Gálvez, A., Huerta, R., Marco, S., (2016). Calibration transfer and drift counteraction in chemical sensor arrays using Direct Standardization Sensors and Actuators B: Chemical 236, 1044-1053
Inherent variability of chemical sensors makes it necessary to calibrate chemical detection systems individually. This shortcoming has traditionally limited usability of systems based on metal oxide gas sensor arrays and prevented mass-production for some applications. Here, aiming at exploring calibration transfer between chemical sensor arrays, we exposed five twin 8-sensor detection units to different concentration levels of ethanol, ethylene, carbon monoxide, or methane. First, we built calibration models using data acquired with a master unit. Second, to explore the transferability of the calibration models, we used Direct Standardization to map the signals of a slave unit to the space of the master unit in calibration. In particular, we evaluated the transferability of the calibration models to other detection units, and within the same unit measuring days apart. Our results show that signals acquired with one unit can be successfully mapped to the space of a reference unit. Hence, calibration models trained with a master unit can be extended to slave units using a reduced number of transfer samples, diminishing thereby calibration costs. Similarly, signals of a sensing unit can be transformed to match sensor behavior in the past to mitigate drift effects. Therefore, the proposed methodology can reduce calibration costs in mass-production and delay recalibrations due to sensor aging. Acquired dataset is made publicly available.
Keywords: Calibration transfer, Chemical sensors, Direct Standardization, Electronic nose, MOX sensors, Public dataset