Predictions For Large Datasets Research Articles

Classification of Pancreatic Ductal Adenocarcinoma Using MALDI Mass Spectrometry Imaging Combined with Neural Networks.

Despite numerous diagnostic and therapeutic advances, pancreatic ductal adenocarcinoma (PDAC) has a high mortality rate, and is the fourth leading cause of cancer death in developing countries. Besides its increasing prevalence, pancreatic malignancies are characterized by poor prognosis. Omics technologies have potential relevance for PDAC assessment but are time-intensive and relatively cost-intensive and limited by tissue heterogeneity. Matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) can obtain spatially distinct peptide-signatures and enables tumor classification within a feasible time with relatively low cost. While MALDI-MSI data sets are inherently large, machine learning methods have the potential to greatly decrease processing time. We present a pilot study investigating the potential of MALDI-MSI in combination with neural networks, for classification of pancreatic ductal adenocarcinoma. Neural-network models were trained to distinguish between pancreatic ductal adenocarcinoma and other pancreatic cancer types. The proposed methods are able to correctly classify the PDAC types with an accuracy of up to 86% and a sensitivity of 82%. This study demonstrates that machine learning tools are able to identify different pancreatic carcinoma from complex MALDI data, enabling fast prediction of large data sets. Our results encourage a more frequent use of MALDI-MSI and machine learning in histopathological studies in the future.

SETH predicts nuances of residue disorder from protein embeddings

Predictions for millions of protein three-dimensional structures are only a few clicks away since the release of AlphaFold2 results for UniProt. However, many proteins have so-called intrinsically disordered regions (IDRs) that do not adopt unique structures in isolation. These IDRs are associated with several diseases, including Alzheimer’s Disease. We showed that three recent disorder measures of AlphaFold2 predictions (pLDDT, “experimentally resolved” prediction and “relative solvent accessibility”) correlated to some extent with IDRs. However, expert methods predict IDRs more reliably by combining complex machine learning models with expert-crafted input features and evolutionary information from multiple sequence alignments (MSAs). MSAs are not always available, especially for IDRs, and are computationally expensive to generate, limiting the scalability of the associated tools. Here, we present the novel method SETH that predicts residue disorder from embeddings generated by the protein Language Model ProtT5, which explicitly only uses single sequences as input. Thereby, our method, relying on a relatively shallow convolutional neural network, outperformed much more complex solutions while being much faster, allowing to create predictions for the human proteome in about 1 hour on a consumer-grade PC with one NVIDIA GeForce RTX 3060. Trained on a continuous disorder scale (CheZOD scores), our method captured subtle variations in disorder, thereby providing important information beyond the binary classification of most methods. High performance paired with speed revealed that SETH’s nuanced disorder predictions for entire proteomes capture aspects of the evolution of organisms. Additionally, SETH could also be used to filter out regions or proteins with probable low-quality AlphaFold2 3D structures to prioritize running the compute-intensive predictions for large data sets. SETH is freely publicly available at: https://github.com/Rostlab/SETH.

22 Accuracy of indirect predictions for large datasets based on prediction error covariance of SNP effects from single-step GBLUP

Abstract With an ever-increasing number of genotyped animals, there is a question of whether to include all genotypes into single-step GBLUP (ssGBLUP) evaluations or to include only genotyped animals with phenotypes and use indirect predictions (IP) for the remaining young genotyped animals. Under ssGBLUP, SNP effects can be backsolved from GEBV, and IP can be calculated as the sum of SNP effects weighted by the gene content. To publish IP, a measure of accuracy that reflects the standard error of prediction, and that is comparable to GEBV accuracy, is needed. Our first objective was to test formulas to compute accuracy of IP by backsolving prediction error covariance (PEC) of GEBV into PEC of SNP effects. The second objective was to investigate the number of genotyped animals needed to obtain robust IP accuracy. Data were provided by the American Angus Association, with 38,000 post-weaning gain phenotypes and 60,000 genotyped animals. Correlations between GEBV and IP were ≥0.99. When all genotyped animals were used for PEC computations, accuracy correlations were also ≥0.99. Additionally, GEBV and IP accuracies were compatible, with both direct inversion of the genomic relationship matrix (G) or using the algorithm for proven and young (APY) to obtain G inverse. As the number of genotyped animals in PEC computations decreased to 15,000, accuracy correlations were still high (≥0.96), but IP accuracies were biased downwards. Indirect prediction accuracy can be successfully obtained from ssGBLUP without running an extra SNP-BLUP evaluation to compute SNP PEC. It is possible to reduce the number of genotyped animals in PEC computations, but accuracies may be slightly underestimated. When the amount of genomic and phenotypic data is large, the polygenic part of GEBV becomes small and IP can be very accurate. Further research is needed to approximate SNP PEC with a large number of genotyped animals.

Osteoporotic hip fracture prediction from risk factors available in administrative claims data - A machine learning approach.

Hip fractures are among the most frequently occurring fragility fractures in older adults, associated with a loss of quality of life, high mortality, and high use of healthcare resources. The aim was to apply the superlearner method to predict osteoporotic hip fractures using administrative claims data and to compare its performance to established methods. We devided claims data of 288,086 individuals aged 65 years and older without care level into a training (80%) and a validation set (20%). Subsequently, we trained a superlearner algorithm that considered both regression and machine learning algorithms (e.g., support vector machines, RUSBoost) on a large set of clinical risk factors. Mean squared error and measures of discrimination and calibration were employed to assess prediction performance. All algorithms used in the analysis showed similar performance with an AUC ranging from 0.66 to 0.72 in the training and 0.65 to 0.70 in the validation set. Superlearner showed good discrimination in the training set but poorer discrimination and calibration in the validation set. The superlearner achieved similar predictive performance compared to the individual algorithms included. Nevertheless, in the presence of non-linearity and complex interactions, this method might be a flexible alternative to be considered for risk prediction in large datasets.

Optimization of the ANOVA Procedure for Support Vector Machines

Feature selection is a powerful tool to identify the important characteristics of data for prediction. Feature selection, therefore, can be a tool for avoiding overfitting, improving prediction accuracy and reducing execution time. The applications of feature selection procedures are particularly important in Support vector machines, which is used for prediction in large datasets. The larger the dataset, the more computationally exhaustive and challenging it is to build a predictive model using the support vector classifier. This paper investigates how the feature selection approach based on the analysis of variance (ANOVA) can be optimized for Support Vector Machines (SVMs) to improve its execution time and accuracy. We introduce new conditions on the SVMs prior to running the ANOVA to optimize the performance of the support vector classifier. We also establish the bootstrap procedure as alternative to cross validation to perform model selection. We run our experiments using popular datasets and compare our results to existing modifications of SVMs with feature selection procedure. We propose a number of ANOVA-SVM modifications which are simple to perform, while at the same time, boost significantly the accuracy and computing time of the SVMs in comparison to existing methods like the Mixed Integer Linear Feature Selection approach.

Postoperative neonatal mortality prediction using superlearning

BackgroundThe variable risks associated with neonatal surgery present a challenge to accurate mortality prediction. We aimed to apply superlearning, an ensemble machine learning method, to the prediction of 30-day neonatal postoperative mortality. Materials and methodsWe included neonates in the 2012-2014 National Surgical Quality Improvement Program Pediatric. Patients treated in 2012-13 were used in model development (n = 6499), and patients treated in 2014 formed the validation sample (n = 3552). Our superlearner algorithm included 14 regression and machine learning algorithms and included all preoperative patient demographic and clinical characteristics, including indicator variables for surgical procedures. Performance was evaluated using mean squared error and measures of discrimination and calibration. ResultsThe superlearner out-performed all individual algorithms with regard to cross-validated mean squared error. It showed excellent discrimination, with an area under the receiver-operating characteristic curve of 0.91 in development and 0.87 in validation. The superlearner showed good calibration in development but not in validation (Cox calibration test P = 0.06 and P < 0.001, respectively). Performance was improved when the superlearner was fit using only variables strongly associated with mortality in bivariate analysis (area under the receiver-operating characteristic curve 0.89, calibration test P = 0.63 in validation). ConclusionsSuperlearning provided improved or equivalent performance compared with individual regression and machine learning algorithms for predicting neonatal surgical mortality. This method should be considered for prediction in large data sets whenever complex mechanisms make parametric modeling assumptions unrealistic.

Variable selection models for genomic selection using whole-genome sequence data and singular value decomposition

BackgroundNon-linear Bayesian genomic prediction models such as BayesA/B/C/R involve iteration and mostly Markov chain Monte Carlo (MCMC) algorithms, which are computationally expensive, especially when whole-genome sequence (WGS) data are analyzed. Singular value decomposition (SVD) of the genotype matrix can facilitate genomic prediction in large datasets, and can be used to estimate marker effects and their prediction error variances (PEV) in a computationally efficient manner. Here, we developed, implemented, and evaluated a direct, non-iterative method for the estimation of marker effects for the BayesC genomic prediction model.MethodsThe BayesC model assumes a priori that markers have normally distributed effects with probability uppi and no effect with probability (1 − uppi ). Marker effects and their PEV are estimated by using SVD and the posterior probability of the marker having a non-zero effect is calculated. These posterior probabilities are used to obtain marker-specific effect variances, which are subsequently used to approximate BayesC estimates of marker effects in a linear model. A computer simulation study was conducted to compare alternative genomic prediction methods, where a single reference generation was used to estimate marker effects, which were subsequently used for 10 generations of forward prediction, for which accuracies were evaluated.ResultsSVD-based posterior probabilities of markers having non-zero effects were generally lower than MCMC-based posterior probabilities, but for some regions the opposite occurred, resulting in clear signals for QTL-rich regions. The accuracies of breeding values estimated using SVD- and MCMC-based BayesC analyses were similar across the 10 generations of forward prediction. For an intermediate number of generations (2 to 5) of forward prediction, accuracies obtained with the BayesC model tended to be slightly higher than accuracies obtained using the best linear unbiased prediction of SNP effects (SNP-BLUP model). When reducing marker density from WGS data to 30 K, SNP-BLUP tended to yield the highest accuracies, at least in the short term.ConclusionsBased on SVD of the genotype matrix, we developed a direct method for the calculation of BayesC estimates of marker effects. Although SVD- and MCMC-based marker effects differed slightly, their prediction accuracies were similar. Assuming that the SVD of the marker genotype matrix is already performed for other reasons (e.g. for SNP-BLUP), computation times for the BayesC predictions were comparable to those of SNP-BLUP.

A fuzzy-based methodology for accurate classification and prediction in large datasets

A fuzzy-based methodology for accurate classification and prediction in large datasets

A practical application of two in silico systems for identification of potentially mutagenic impurities

The International Conference on Harmonization (ICH) M7 guidance for the assessment and control of DNA reactive impurities in pharmaceutical products includes the use of in silico prediction systems as part of the hazard identification and risk assessment strategy. This is the first internationally agreed guidance document to include the use of these types of approaches. The guideline requires the use of two complementary approaches, an expert rule-based method and a statistical algorithm. In addition, the guidance states that the output from these computer-based assessments can be reviewed using expert knowledge to provide additional support or resolve conflicting predictions. This approach is designed to maximize the sensitivity for correctly identifying DNA reactive compounds while providing a framework to reduce the number of compounds that need to be synthesized, purified and subsequently tested in an Ames assay. Using a data set of 801 chemicals and pharmaceutical intermediates, we have examined the relative predictive performances of some popular commercial in silico systems that are in common use across the pharmaceutical industry. The overall accuracy of each of these systems was fairly comparable ranging from 68% to 73%; however, the sensitivity of each system (i.e. how many Ames positive compounds are correctly identified) varied much more dramatically from 48% to 68%. We have explored how these systems can be combined under the ICH M7 guidance to enhance the detection of DNA reactive molecules. Finally, using four smaller sets of molecules, we have explored the value of expert knowledge in the review process, especially in cases where the two systems disagreed on their predictions, and the need for care when evaluating the predictions for large data sets.

Qualified Predictions for Large Data Sets

Qualified Predictions for Large Data Sets