Data-driven characterization of molecular phenotypes across heterogeneous sample collections.

Existing large gene expression data repositories hold enormous potential to elucidate disease mechanisms, characterize changes in cellular pathways, and to stratify patients based on molecular profiles. To achieve this goal, integrative resources and tools are needed that allow comparison of results across datasets and data types. We propose an intuitive approach for data-driven stratifications of molecular profiles and benchmark our methodology using the dimensionality reduction algorithm t-distributed stochastic neighbor embedding (t-SNE) with multi-study and multi-platform data on hematological malignancies. Our approach enables assessing the contribution of biological versus technical variation to sample clustering, direct incorporation of additional datasets to the same low dimensional representation, comparison of molecular disease subtypes identified from separate t-SNE representations, and characterization of the obtained clusters based on pathway databases and additional data. In this manner, we performed an integrative analysis across multi-omics acute myeloid leukemia studies. Our approach indicated new molecular subtypes with differential survival and drug responsiveness among samples lacking fusion genes, including a novel myelodysplastic syndrome-like cluster and a cluster characterized with CEBPA mutations and differential activity of the S-adenosylmethionine-dependent DNA methylation pathway. In summary, integration across multiple studies can help to identify novel molecular disease subtypes and generate insight into disease biology.

On the limits of automatic speaker verification: Explaining degraded recognizer scores through acoustic changes resulting from voice disguise.

In speaker verification research, objective performance benchmarking of listeners and automatic speaker verification (ASV) systems are of key importance in understanding the limits of speaker recognition. While the adoption of common data and metrics has been instrumental to progress in ASV, there are two major shortcomings. First, the utterances lack intentional voice changes imposed by the speaker. Second, the standard evaluation metrics focus on average performance across all speakers and trials. As a result, a knowledge gap remains in how the acoustic changes impact recognition performance at the level of individual speakers. This paper addresses the limits of speaker recognition in ASV systems under voice disguise using a linear mixed effects model to analyze the impact of change in long-term statistics of selected features (formants F1-F4, the bandwidths B1-B4, F0, and speaking rate) to ASV log-likelihood ratio (LLR) score. The correlations between the proposed predictive model and the LLR scores are 0.72 for females and 0.81 for male speakers. As a whole, the difference in long-term F0 between enrollment and test utterances was found to be the individually most detrimental factor, even if the ASV system uses only spectral, rather than prosodic, features.

Inferring personal intake recommendations of phosphorous and potassium for end-stage renal failure patients by simulating with Bayesian hierarchical multivariate model.

Most end-stage renal disease (ESRD) patients face a risk of malnutrition, partly due to dietary restrictions on phosphorous and, in some cases, potassium intake. These restrictions aim to regulate plasma phosphate and potassium concentrations and prevent the adverse effects of hyperphosphatemia or hyperkalemia. However, individual responses to nutrition are known to vary, highlighting the need for personalized recommendations rather than relying solely on general guidelines. In this study, our objective was to develop a Bayesian hierarchical multivariate model that estimates the individual effects of nutrients on plasma concentrations and to present a recommendation algorithm that utilizes this model to infer personalized dietary intakes capable of achieving normal ranges for all considered concentrations. Considering the limited research on the reactions of ESRD patients, we collected dietary intake data and corresponding laboratory analyses from a cohort of 37 patients. The collected data were used to estimate the common hierarchical model, from which personalized models of the patients' diets and individual reactions were extracted. The application of our recommendation algorithm revealed substantial variations in phosphorus and potassium intakes recommended for each patient. These personalized recommendations deviate from the general guidelines, suggesting that a notably richer diet may be proposed for certain patients to mitigate the risk of malnutrition. Furthermore, all the participants underwent either hospital, home, or peritoneal dialysis treatments. We explored the impact of treatment type on nutritional reactions by incorporating it as a nested level in the hierarchical model. Remarkably, this incorporation improved the fit of the nutritional effect model by a notable reduction in the normalized root mean square error (NRMSE) from 0.078 to 0.003. These findings highlight the potential for personalized dietary modifications to optimize nutritional status, enhance patient outcomes, and mitigate the risk of malnutrition in the ESRD population.

Mixed-effect Bayesian network reveals personal effects of nutrition.

Nutrition experts know by their experience that people can react very differently to the same nutrition. If we could systematically quantify these differences, it would enable more personal dietary understanding and guidance. This work proposes a mixed-effect Bayesian network as a method for modeling the multivariate system of nutrition effects. Estimation of this network reveals a system of both population-wide and personal correlations between nutrients and their biological responses. Fully Bayesian estimation in the method allows managing the uncertainty in parameters and incorporating the existing nutritional knowledge into the model. The method is evaluated by modeling data from a dietary intervention study, called Sysdimet, which contains personal observations from food records and the corresponding fasting concentrations of blood cholesterol, glucose, and insulin. The model's usefulness in nutritional guidance is evaluated by predicting personally if a given diet increases or decreases future levels of concentrations. The proposed method is shown to be comparable with the well-performing Extreme Gradient Boosting (XGBoost) decision tree method in classifying the directions of concentration increases and decreases. In addition to classification, we can also predict the precise concentration level and use the biologically interpretable model parameters to understand what personal effects contribute to the concentration. We found considerable personal differences in the contributing nutrients, and while these nutritional effects are previously known at a population level, recognizing their personal differences would result in more accurate estimates and more effective nutritional guidance.

Semisupervised Generative Autoencoder for Single-Cell Data.

Single-cell transcriptomics offers a tool to study the diversity of cell phenotypes through snapshots of the abundance of mRNA in individual cells. Often there is additional information available besides the single-cell gene expression counts, such as bulk transcriptome data from the same tissue, or quantification of surface protein levels from the same cells. In this study, we propose models based on the Bayesian deep learning approach, where protein quantification, available as CITE-seq counts, from the same cells is used to constrain the learning process, thus forming a SemI-SUpervised generative Autoencoder (SISUA) model. The generative model is based on the deep variational autoencoder (VAE) neural network architecture.

Hemap: An Interactive Online Resource for Characterizing Molecular Phenotypes across Hematologic Malignancies.

Large collections of genome-wide data can facilitate the characterization of disease states and subtypes, permitting pan-cancer analysis of molecular phenotypes and evaluation of disease context for new therapeutic approaches. We analyzed 9,544 transcriptomes from more than 30 hematologic malignancies, normal blood cell types, and cell lines, and showed that disease types could be stratified in a data-driven manner. We then identified cluster-specific pathway activity, new biomarkers, and in silico drug target prioritization through interrogation of drug target databases. Using known vulnerabilities and available drug screens, we highlighted the importance of integrating molecular phenotype with drug target expression for in silico prediction of drug responsiveness. Our analysis implicated BCL2 expression level as an important indicator of venetoclax responsiveness and provided a rationale for its targeting in specific leukemia subtypes and multiple myeloma, linked several polycomb group proteins that could be targeted by small molecules (SFMBT1, CBX7, and EZH1) with chronic lymphocytic leukemia, and supported CDK6 as a disease-specific target in acute myeloid leukemia. Through integration with proteomics data, we characterized target protein expression for pre-B leukemia immunotherapy candidates, including DPEP1. These molecular data can be explored using our publicly available interactive resource, Hemap, for expediting therapeutic innovations in hematologic malignancies. SIGNIFICANCE: This study describes a data resource for researching derailed cellular pathways and candidate drug targets across hematologic malignancies.

Fast agglomerative clustering using a k-nearest neighbor graph.

We propose a fast agglomerative clustering method using an approximate nearest neighbor graph for reducing the number of distance calculations. The time complexity of the algorithm is improved from O(tauN2) to O(tauNlogN) at the cost of a slight increase in distortion; here, tau denotes the number of nearest neighbor updates required at each iteration. According to the experiments, a relatively small neighborhood size is sufficient to maintain the quality close to that of the full search.