The Promises and Perils of MS Machine Learning
Panel, ASMS Asilomar 2023, Monterey, CA
This panel discussed topics related to machine learning applications in mass spectrometry(MS)-based proteomics.
Talks and presentations
Single cell mass spectrometry (MS)-based proteomics coupled with suitable experimental design enables the application of causal inference methods to observational experiments.
Talk, iSCMS 2023, Provo, UT
This talk explored the integration of new single cell MS-based proteomics methods and causal inference. We argued that single cell MS-based proteomics enables estimating the effect of perturbations from observational studies. These methods were shown on a simulation and real world single cell experiment.
Computational causal inference methods and single cell proteomics enable the estimation of interventions from purely observational experiments
Talk, ASMS Asilomar 2023, Monterey, CA
This talk explored the integration of new single cell MS-based proteomics methods and causal inference. We argued that single cell MS-based proteomics enables estimating the effect of perturbations from observational studies. These methods were shown on a simulation and real world single cell experiment.
Bayesian statistical modeling reveals missing value mechanisms in label-free Mass Spectrometry-based proteomics experiments
Talk, ASMS 2023, Houston, TX
This talk reviewed bayesian statistical modeling strategies for upstream data processing of MS-based proteomics experiments. This included missing value imputation and summarization of peptide ions to the protein-level. These processing steps were performed with a bayesian model which allowed error propogation at each step. The final summarized values were used for differential analysis in a error-in-variance model.
Data Independent Acquisition: After the Acquisition
Panel, ASMS 2023, Houston, TX
This panel discussed topics related to mass spectrometry(MS)-based proteomics experiments using data independent acquisition (DIA). Specifically, it focused on what to do with the data resulting from these experiments. This included indentification and quantificaiton of mass spectra, as well as post id and quantificaiton processing of the data.
Analysis of Quantitative MS-based Proteomics Experiments using MSstats and MSstatsTMT
Talk, EMERGE webinar series - ReSyn Biosciences, Online
Recent advances in the techniques and technologies used in Mass Spectrometry (MS)-based proteomics have greatly increased the variety and complexity of experimental designs in the field. Experiments can differ in the labeling method (label-free vs tandem mass tag), acquisition type (DDA/DIA/SRM/PRM), biological question of interest, and differing numbers of conditions and replicates. Statistical methods used to analyze these experiments need to be flexible enough to fit these design complexities, while being robust enough to not overfit to one specific design. The MSstats family of R packages is widely used to analyze the results of MS-based proteomics experiments and has been shown to outperform other methods on a variety of experimental designs. In this webinar we will review the core workflow and methods behind MSstats and its extensions, highlighting critical choices that need to be made in each step of the analysis. We will review how different analysis choices affect the final protein-level conclusions made on the experiment and discuss how to avoid potential analysis pitfalls. After reviewing the methods in MSstats, we will present a hands-on session using MSstatsShiny, where participants are provided with two datasets (label-free DIA and TMT-DDA) and are invited to follow along.
Single-cell mass spectrometry-based proteomics enables causal inference in observational studies
Poster, US HUPO 2023, Chicago, IL
Single cell proteomics greatly increases the number of replicates and cellular resolution in MS-based proteomic experiments. However, these experiments are expensive (especially when targeting perturbations). Causal inference methods, which are typically challenging to apply to traditional bulk-MS due to the lack of replicates, allow us to estimate the impact of perturbations from purely observational data. We propose a method and workflow for predicting the effect of interventions in observational single cell MS experiments and apply the workflow to a recent observational single cell experiment.
MSstatsShiny: A Multipurpose UI for Reproducible Analysis of Quantitative Proteomic Experiments
Poster, ASMS 2022, Minneapolis, MN
Quantitative mass spectrometry proteomic experiments require robust analytic and modeling techniques to ensure the results are correctly interpreted. There are a variety of different tools that can assist researchers in analysis, however they are generally only suited for one type of experimental design and are usually implemented in coding packages. To address these challenges, we have created the UI MSstats-Shiny, an RShiny based UI integrated with the packages Msstats, MSstatsTMT, and MSstatsPTM providing all users an end-to-end pipeline that can analyze a variety of experimental designs.
Explorations of causal probabilistic programming approaches for rule-based models of biological signaling pathways
Poster, Prob Prog 2021, Online
Rule-based models handle the complexity of biological signaling pathways. Biological signalling pathways are complex systems that underlie many cellular process and whose dysregulation is the source of many morbidities. To address the combinatorial complexity of interactions, patterns of transitions between model states can be compactly represented as probabilistic events using rule-based models.
MSstatsPTM: an R/Bioconductor software for detecting quantitative changes in post-translational modifications
Poster, US HUPO 2021, Online
The scientific community widely utilizes mass spectrometry (MS)-based proteomics to quantify the abundance of proteins and their post-translational modifications (PTMs). Experiments targeting PTMs face several specific challenges. These include the low abundance, few representative peptides, and convolution with abundance changes in the overall protein expression. Due to these challenges, a robust approach to estimate relative changes in PTMs should combine PTM sites over several peptides, replicates in multiple conditions, and consider sources of confounding present in the experiment. We propose a general statistical model and workflow that is both reproducible and comprehensive. The method measures PTM and protein abundance by summarizing intensities through Tukey’s median polish method. Then a model based on the family of linear mixed-effects models is fit. Finally, the PTM abundances are adjusted to remove variance from changes in the overall protein. The package can handle a diverse range of acquisition types, including label-free, DDA, DIA, and TMT. We implement this model in the free and open-source R package MSstatsPTM and available at the links below.