In deep learning for drug discovery, molecular representations are often based on sequences, known as SMILES, which allow for straightforward implementation of natural language processing methodologies, one being the sequence-to-sequence autoencoder. However, we observe that training an autoencoder solely on SMILES is insufficient to learn molecular representations that are semantically meaningful, where semantics are specified by the structural (graph-to-graph) similarities between molecules. We demonstrate by example that SMILES-based autoencoders may map structurally similar molecules to distant codes, resulting in an incoherent latent space that does not necessarily respect the semantic similarities between molecules. To address this shortcoming we propose Semantically-Aware Latent Space Autoencoder (SALSA) for molecular representations: a SMILES-based transformer autoencoder modified with a contrastive task aimed at learning graph-to-graph similarities between molecules. To accomplish this, we develop a novel dataset comprised of sets of structurally similar molecules and opt for a supervised contrastive loss that is able to incorporate full sets of positive samples. We evaluate semantic awareness of SALSA representations by comparing to its ablated counterparts, and show empirically that SALSA learns representations that maintain 1) structural awareness, 2) physicochemical awareness, 3) biological awareness, and 4) semantic continuity.

Protein kinases play a vital role in a wide range of cellular processes, and compounds that inhibit kinase activity emerging as a primary focus for targeted therapy development, especially in cancer. Consequently, efforts to characterize the behavior of kinases in response to inhibitor treatment, as well as downstream cellular responses, have been performed at increasingly large scales. Previous work with smaller datasets have used baseline profiling of cell lines and limited kinome profiling data to attempt to predict small molecule effects on cell viability, but these efforts did not use multi-dose kinase profiles and achieved low accuracy with very limited external validation. This work focuses on two large-scale primary data types, kinase inhibitor profiles and gene expression, to predict the results of cell viability screening. We describe the process by which we combined these data sets, examined their properties in relation to cell viability and finally developed a set of computational models that achieve a reasonably high prediction accuracy (R2 of 0.78 and RMSE of 0.154). Using these models, we identified a set of kinases, several of which are understudied, that are strongly influential in the cell viability prediction models. In addition, we also tested to see if a wider range of multiomics data sets could improve the model results and found that proteomic kinase inhibitor profiles were the single most informative data type. Finally, we validated a small subset of the model predictions in several triple-negative and HER2 positive breast cancer cell lines demonstrating that the model performs well with compounds and cell lines that were not included in the training data set. Overall, this result demonstrates that generic knowledge of the kinome is predictive of very specific cell phenotypes, and has the potential to be integrated into targeted therapy development pipelines.

MOTIVATION: Kinase-catalyzed phosphorylation of proteins forms the backbone of signal transduction within the cell, enabling the coordination of numerous processes such as the cell cycle, apoptosis, and differentiation. Although on the order of 105 phosphorylation events have been described, we know the specific kinase performing these functions for <5% of cases. The ability to predict which kinases initiate specific individual phosphorylation events has the potential to greatly enhance the design of downstream experimental studies, while simultaneously creating a preliminary map of the broader phosphorylation network that controls cellular signaling. RESULTS: We describe Embedding-based multi-label prediction of phosphorylation events (EMBER), a deep learning method that integrates kinase phylogenetic information and motif-dissimilarity information into a multi-label classification model for the prediction of kinase-motif phosphorylation events. Unlike previous deep learning methods that perform single-label classification, we restate the task of kinase-motif phosphorylation prediction as a multi-label problem, allowing us to train a single unified model rather than a separate model for each of the 134 kinase families. We utilize a Siamese neural network to generate novel vector representations, or an embedding, of peptide motif sequences, and we compare our novel embedding to a previously proposed peptide embedding. Our motif vector representations are used, along with one-hot encoded motif sequences, as input to a classification neural network while also leveraging kinase phylogenetic relationships into our model via a kinase phylogeny-weighted loss function. Results suggest that this approach holds significant promise for improving the known map of phosphorylation relationships that underlie kinome signaling. AVAILABILITY AND IMPLEMENTATION: The data and code underlying this article are available in a GitHub repository at https://github.com/gomezlab/EMBER. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Timely, and sometimes rapid, metabolic adaptation to changes in food supply is critical for survival as an organism moves from the fasted to the fed state, and vice versa. These transitions necessitate major metabolic changes to maintain energy homeostasis as the source of blood glucose moves away from ingested carbohydrates, through hepatic glycogen stores, towards gluconeogenesis. The integration of hepatic glycogen regulation with extra-hepatic energetics is a key aspect of these adaptive mechanisms. Here we use computational modeling to explore hepatic glycogen regulation under fed and fasting conditions in the context of a whole-body model. The model was validated against previous experimental results concerning glycogen phosphorylase a (active) and glycogen synthase a dynamics. The model qualitatively reproduced physiological changes that occur during transition from the fed to the fasted state. Analysis of the model reveals a critical role for the inhibition of glycogen synthase phosphatase by glycogen phosphorylase a. This negative regulation leads to high levels of glycogen synthase activity during fasting conditions, which in turn increases substrate (futile) cycling, priming the system for a rapid response once an external source of glucose is restored. This work demonstrates that a mechanistic understanding of the design principles used by metabolic control circuits to maintain homeostasis can benefit from the incorporation of mathematical descriptions of these networks into "whole-body" contextual models that mimic in vivo conditions.

BACKGROUND: Nuclear magnetic resonance spectroscopy is one of the primary tools in metabolomics analyses, where it is used to track and quantify changes in metabolite concentrations or profiles in response to perturbation through disease, toxicants or drugs. The spectra generated through such analyses are typically confounded by noise of various types, obscuring the signals and hindering downstream statistical analysis. Such issues are becoming increasingly significant as greater numbers of large-scale systems or longitudinal studies are being performed, in which many spectra from different conditions need to be compared simultaneously. RESULTS: We describe a novel approach, termed Progressive Consensus Alignment of Nmr Spectra (PCANS), for the alignment of NMR spectra. Through the progressive integration of many pairwise comparisons, this approach generates a single consensus spectrum as an output that is then used to adjust the chemical shift positions of the peaks from the original input spectra to their final aligned positions. We characterize the performance of PCANS by aligning simulated NMR spectra, which have been provided with user-defined amounts of chemical shift variation as well as inter-group differences as would be observed in control-treatment applications. Moreover, we demonstrate how our method provides better performance than either template-based alignment or binning. Finally, we further evaluate this approach in the alignment of real mouse urine spectra and demonstrate its ability to improve downstream PCA and PLS analyses. CONCLUSIONS: By avoiding the use of a template or reference spectrum, PCANS allows for the creation of a consensus spectrum that enhances the signals within the spectra while maintaining sample-specific features. This approach is of greatest benefit when complex samples are being analyzed and where it is expected that there will be spectral features unique and/or strongly different between subgroups within the samples. Furthermore, this approach can be potentially applied to the alignment of any data having spectra-like properties.

BACKGROUND: The understanding of evolutionary relationships is a fundamental aspect of modern biology, with the phylogenetic tree being a primary tool for describing these associations. However, comparison of trees for the purpose of assessing similarity and the quantification of various biological processes remains a significant challenge. RESULTS: We describe a novel approach for the comparison of phylogenetic distance information based on the alignment of representative high-dimensional embeddings (xCEED: Comparison of Embedded Evolutionary Distances). The xCEED methodology, which utilizes multidimensional scaling and Procrustes-related superimposition approaches, provides the ability to measure the global similarity between trees as well as incongruities between them. We demonstrate the application of this approach to the prediction of coevolving protein interactions and demonstrate its improved performance over the mirrortree, tol-mirrortree, phylogenetic vector projection, and partial correlation approaches. Furthermore, we show its applicability to both the detection of horizontal gene transfer events as well as its potential use in the prediction of interaction specificity between a pair of multigene families. CONCLUSIONS: These approaches provide additional tools for the study of phylogenetic trees and associated evolutionary processes. Source code is available at http://gomezlab.bme.unc.edu/tools.

In order to understand the molecular machinery of the cell, we need to know about the multitude of protein-protein interactions that allow the cell to function. High-throughput technologies provide some data about these interactions, but so far that data is fairly noisy. Therefore, computational techniques for predicting protein-protein interactions could be of significant value. One approach to predicting interactions in silico is to produce from first principles a detailed model of a candidate interaction. We take an alternative approach, employing a relatively simple model that learns dynamically from a large collection of data. In this work, we describe an attraction-repulsion model, in which the interaction between a pair of proteins is represented as the sum of attractive and repulsive forces associated with small, domain- or motif-sized features along the length of each protein. The model is discriminative, learning simultaneously from known interactions and from pairs of proteins that are known (or suspected) not to interact. The model is efficient to compute and scales well to very large collections of data. In a cross-validated comparison using known yeast interactions, the attraction-repulsion method performs better than several competing techniques.

Knowledge on interactions between molecules in living cells is indispensable for theoretical analysis and practical applications in modern genomics and molecular biology. Building such networks relies on the assumption that the correct molecular interactions are known or can be identified by reading a few research articles. However, this assumption does not necessarily hold, as truth is rather an emerging property based on many potentially conflicting facts. This paper explores the processes of knowledge generation and publishing in the molecular biology literature using modelling and analysis of real molecular interaction data. The data analysed in this article were automatically extracted from 50000 research articles in molecular biology using a computer system called GeneWays containing a natural language processing module. The paper indicates that truthfulness of statements is associated in the minds of scientists with the relative importance (connectedness) of substances under study, revealing a potential selection bias in the reporting of research results. Aiming at understanding the statistical properties of the life cycle of biological facts reported in research articles, we formulate a stochastic model describing generation and propagation of knowledge about molecular interactions through scientific publications. We hope that in the future such a model can be useful for automatically producing consensus views of molecular interaction data.

MOTIVATION: Current growth in the field of genomics has provided a number of exciting approaches to the modeling of evolutionary mechanisms within the genome. Separately, dynamical and statistical analyses of networks such as the World Wide Web and the social interactions existing between humans have shown that these networks can exhibit common fractal properties-including the property of being scale-free. This work attempts to bridge these two fields and demonstrate that the fractal properties of molecular networks are linked to the fractal properties of their underlying genomes. RESULTS: We suggest a stochastic model capable of describing the evolutionary growth of metabolic or signal-transduction networks. This model generates networks that share important statistical properties (so-called scale-free behavior) with real molecular networks. In particular, the frequency of vertices connected to exactly k other vertices follows a power-law distribution. The shape of this distribution remains invariant to changes in network scale: a small subgraph has the same distribution as the complete graph from which it is derived. Furthermore, the model correctly predicts that the frequencies of distinct DNA and protein domains also follow a power-law distribution. Finally, the model leads to a simple equation linking the total number of different DNA and protein domains in a genome with both the total number of genes and the overall network topology. AVAILABILITY: MatLab (MathWorks, Inc.) programs described in this manuscript are available on request from the authors. CONTACT: [email protected].

MOTIVATION: In order to aid in hypothesis-driven experimental gene discovery, we are designing a computer application for the automatic retrieval of signal transduction data from electronic versions of scientific publications using natural language processing (NLP) techniques, as well as for visualizing and editing representations of regulatory systems. These systems describe both signal transduction and biochemical pathways within complex multicellular organisms, yeast, and bacteria. This computer application in turn requires the development of a domain-specific ontology, or knowledge model. RESULTS: We introduce an ontological model for the representation of biological knowledge related to regulatory networks in vertebrates. We outline a taxonomy of the concepts, define their 'whole-to-part' relationships, describe the properties of major concepts, and outline a set of the most important axioms. The ontology is partially realized in a computer system designed to aid researchers in biology and medicine in visualizing and editing a representation of a signal transduction system.