Posters - Schedules

Interest in modified RNA has spiked in recent years with the growing application of Oligonucleotides as therapeutics. This intersects with the realization that RNA modifications in vivo play a huge role in diverse cellular processes. As we have shown previously the NucleicAcidSearchEngine (a part of OpenMS) allows modification identification and localization across a wide variety of different types of nucleotides. As the leading oligonucleotide search engine, we have continued to strive to improve our ability to identify larger nucleic acids. In this poster we demonstrate synergy between the NucleicAcidSearchEngine and another new OpenMS tool FLASHDeconv, to further improve our ability to de-charge large nucleic acids. We compare NASE's built in deconvolution and decharging features to those contained in FLASHDeconv and demonstrate that even with the improvements discussed in our poster at ASMS FLASHDeconv offers complementary benefits for identifying larger oligonucleotides.

Mass spectrometry (MS) is a powerful and unbiased tool for studying COVID-19. Although comorbidities are widely considered risk factors of COVID-19 severity, an MS-based approach to understand the association between comorbidities and patient outcomes is hindered by the extensive patient heterogeneity and the limited sample size typical of such studies. We conducted data-independent acquisition MS-based proteomics analysis of plasma from 88 hospitalized COVID-19 patients with and without comorbidities. To identify protein signatures of comorbidities in the context of COVID-19, we first fitted robust linear models to each peptide, adjusting variables including age, gender, and race. Next, we evaluated whether a comorbidity caused coordinated changes in peptide abundance for each protein by repurposing the “CAMERA (Correlation Adjusted MEan RAnk gene set test)” function in the LIMMA R package for peptide set testing. Finally, we applied the same approach to identify protein signatures linked to patient survival and searched for overlaps between comorbidity and prognostic signatures. Seventeen, 21,16, and 23 proteins associated with hypertension, diabetes, obesity, and renal disease, respectively, coincided with the prognostic protein signature, reinforcing previous findings of comorbidities as risk factors. Importantly, proteomic differences between comorbidities indicated that each comorbidity may contribute to COVID-19 severity in a unique manner.

The combination of stable isotope labeling with proteomics analysis enables tracking of nutrient flows into specific taxa and active metabolic processes. However, few bioinformatics solutions are available for capturing differential protein labeling within and between organisms in microbial communities. Here we build on our previous pipeline for quantification of peptide-specific label incorporation by (1) adapting it for use with alternative search algorithms and benchmarking performance, and (2) evaluating the impact of variable modifications during peptide-spectrum matching. Using bacterial culture data with known atom% 15N labeling, we found that peptide identifications declined with increasing isotopic label, regardless of search algorithm or variable modifications allowed. While the original pipeline development with SequestHT (ProteomeDiscoverer) showed the highest identification rates with highly accurate isotope enrichment estimates, MSGF+ (in OpenMS) appears to be a comparable freely available alternative, albeit with fewer peptide identifications above ~20 atom% 15N. Little variation in peptide identification rates between 5 - 20 variable modifications per peptide suggests that fewer variable modifications can be used to reduce compute time without substantial impact on the results. By extending our pipeline to freely available search algorithms, this work broadens accessibility of a unique bioinformatic solution for tracking assimilation of labeled substrates within metaproteomes.

De novo peptide sequencing (DNPS), determining the peptide amino acid sequence from a tandem mass spectrum, could make proteomics amenable for applications including genotyping and metagenomics. However, DNPS is highly ambiguous with state-of-the-art performance having poor recall at high precision. Here we propose three innovations that, when combined, improve DNPS but can also be used individually. First, we consider DNPS as a bin classification problem: whether a discretized m/z value (bin) of a spectrum belongs to a particular ion series. Second, we introduce an amino-acid-gapped convolution layer that is designed to connect distant bins to form consistent ion series. Third, we introduce a fitness function to evaluate how well a candidate peptide matches a given spectrum by training a model estimating the number of single amino acid editions to the correct peptide. Bin classification and the fitness function leverage the peptide-to-spectrum predictor Prosit. The bin classification model yielded high precision-recall of bin classes and the fitness function precisely evaluates any peptide-spectrum match. We combined the methods in a genetic algorithm. Initial results of the genetic algorithm on a human cell line dataset increased the recall by 19.3% at 90% precision and the overall recall improved by 35.6%.

Mass spectrometry-based quantitative proteomics is rapidly gaining importance in biological and clinical research. Among the variety of quantitative proteomics techniques available, label-free quantification (LFQ) is one of the most commonly used applications for protein quantification across biological samples or conditions. However, the process of analyzing data generated from such proteomics experiments can be often complicated due to the complex and diverse nature of data sets. There is still a great need for specialized tools that can simplify the interpretation and downstream statistical analysis of these complex data sets. Here, we present promor, a user-friendly analytical tool that harnesses the power of R statistical environment and machine learning to perform quality control, visualization, differential expression analysis, and modeling of proteomics data. Compared to existing analytical tools, promor provides a wealth of new functionalities in terms of handling technical replicates in proteomics data, imputation of missing data using a variety of imputation methods, and evaluation of predictive performance of proteomics-based candidate biomarkers using machine learning approaches.

Spectral library search remains one of the most reliable mass spectrometry analyses to determine the chemical composition of a sample. Candidate compounds are recommended by matching the test spectrum against reference spectra. Understanding how a candidate molecule contributes to the mass spectrum increases confidence in metabolite assignment. We present PEEK, an algorithm that leverages the hierarchical structural relationships of known compounds from ontologies to provide explainability to mass fragments. PEEK traverses the ontology to identify entries that are structurally related to the candidate structure and subsequently assigns each spectral peak with a structure. Of the 10,032,347 peaks from the 305,123 [M+H]+ NIST 20 spectra, 1,691,882 (16.86%) peaks were labeled by both PEEK with a chemical structure and NIST with a chemical formula. A comparison of the chemical formulae shows that 930,789 peaks (55.02%) have matching formulae from both sources while chemical formulae from 75,6012 (44.68%) peaks differ by up to only two hydrogen atoms. The discrepancy can be explained by hydrogen rearrangement and/or adduct formation. We demonstrated that PEEK labels [M+H]+ spectral peaks with remarkable accuracy, making it a promising tool to complement current spectral search algorithms.

Native top-down mass spectrometry (MS)-based proteomics is a powerful method for the comprehensive characterization of proteoforms and intact protein complexes in their native state. However, practitioners of native MS are challenged by the complex datasets generated by native top-down MS experiments and a lack of software tools designed to cope with problems unique to native MS. Herein, we present MASH Native, a comprehensive software application for native top-down proteomics. MASH Native is a multithreaded Windows application implemented in C# using the .NET framework and provides various functionalities for native top-down MS data interpretation and processing through the incorporation of many deconvolution methods including UniDec, multiple searching algorithm support, spectral averaging, internal fragmentation searching and proteoform quantitation. Importantly, MASH Native is a freely available software package and can process datasets from various vendor formats while still retaining MASH Explorer’s capability to process denatured top-down proteomics data. With the support of multiple file formats, the integration of numerous analysis tools, and the additional navigation, validation, and manual search functionalities; MASH Native is a universal, comprehensive, user-friendly, and vital tool for analyzing any native or denatured top-down MS experimental data.

Diabetes mellitus prevalence has reached pandemic proportions, and diabetic cardiomyopathy (DC) is a significant and frequently (60%) associated complication. This condition causes early diastolic and late systolic dysfunction and makes patients prone to heart failure. Alteration of myofilament site-specific phosphorylation stoichiometry is evident in experimental and human failing hearts. Similarly, altered phosphorylation levels of individual myofilament proteins have been associated with DC. Another important post-translational modification (PTM) associated with diabetes is O-GlcNAcylation which is linked to glucose metabolism and which, like phosphorylation, modifies both serine and threonine residues. In order to find the missing link between hyperglycemia and abnormal cardiac function in diabetic, we generated global myofilament site-specific O-GlcNAcylation and Phosphorylation data in normal and type 2 diabetic hearts of mouse models. In this talk, we present the result of our investigation in the interplay between OGlcNAcylation and phosphorylation in the diabetic cardiomyopathy. Our results show that there is significantly high correlation between the fold change of phosphorylation and O-GlcNAcylation of the same residues. Moreover, we show that the closely positioned intra-protein residues have higher correlation between the phosphorylation and OGlcNAcylation, and as the sites get far away from each other, their correlation reduces.

Kinase activity inference based on measurements of known substrates is routinely applied to mass spectroscopy-based phosphoproteomics data, but there remains no consensus around which inference method to use. We established a benchmark using CPTAC phosphoproteomics data from 838 tumors across eight cancer types. For each kinase with a known activating site, we selected samples with high and low levels of the site to form a positive group and a negative group, respectively, and then compared eight methods for their ability to distinguish the two groups. With AUROC=0.75, a simple method computing the mean of known kinase substrates in each sample outperformed all competing methods. Furthermore, adding substrates predicted by NetworKIN into the mean computation increased the AUROC to from 0.79 to 0.84. Taking advantage of matched genomics data for these samples, we systematically investigated the association of genetic aberrations of tumor suppressors and oncogenes to inferred kinase activity. EGFR activity was significantly higher in tumors with EGFR amplification and elevated CDK6 activity was associated with loss of CDKN2A in glioblastoma and lung cancers, highlighting the potential for using accurate kinase activity inference to connect genetic aberrations of unknown functional consequence to altered signaling in human tumors.

Untargeted metabolomics experiments rely on spectral libraries for structure
annotation, but these libraries are vastly incomplete; in-silico methods
search in structure databases, allowing us to overcome this limitation. The best-performing in-silico methods
use machine learning to predict a molecular fingerprint from tandem mass spectra, then use the predicted fingerprint to search in a molecular structure database. Predicted molecular fingerprints are also of great interest for compound class annotation, de novo structure elucidation, and other tasks. So far, Kernel support vector machines are the best tool for fingerprint prediction. However, they cannot be trained on all publicly available reference spectra because their training time scales cubically with the number of training data.
Here, we use the Nystrom approximation to transform the kernel into a linear feature map. We evaluate two methods that use this feature map as input: a linear SVM and a deep neural network.
For evaluation we use a cross-validated dataset of 156,017 compounds and three independent datasets with 1,734 compounds. We show that the combination of kernel method and deep neural network outperforms the kernel support vector machine, which is the current gold-standard, as well as a deep neural network on tandem mass spectra on all evaluation datasets.