NIH Proteomics Interest Group

Link to Upcoming Meetings

Link to Meeting Archives

Link to Related Links

Link to NIH Labs

Link Fellows Forum Info

Link to Join SIG

Line















ProtIG is an NIH Special Interest Group (SIG) that organizes seminars and workshops in relevant areas of proteomics, including talks on separation and protein identification methods, determination of post-translational modifications, protein-protein interactions, and bioinformatics and data management. A monthly seminar series is usually held at 10 am on the Second Thursday of each month (always check the Mtgs/Seminars button on this page for these and other PROTIG announced meetings). To receive email announcements of ProtIG events, join the listserv (Join the SIG button on this page)

November ProtIG Seminar
Please note the unusual time and location
November 8th, 2018
10:00 am - 11:00am
Building 50, NIH Campus
Room 1227/1233 (Front Lobby Conference Room)
Samuel Katz
Graduate Student
Oxford-Cambridge Scholars Program, NIAID


Nominated by Iain Fraser, Ph.D

"Beyond Top Scores and Known Pathways: A TRIAGE approach to more robustly interrogate the depth of -omic datasets"

Advancements in genomic and proteomic screening technologies have made it possible to generate extensive lists of targets implicated in the regulation of various signal transduction pathways. However, the subsequent validation of candidates generated by these studies is significantly hindered by the paired challenges of correcting for the noise intrinsic to the scale of these results. A diverse set of bioinformatic solutions has already been created to address this challenge, such as enrichment analysis to reduce the false positive rate and network analysis to decrease the false negative rate. Despite the complimentary corrections of these two methods there is an absence of an integrative model for combining their divergent outputs such that the combinatorial benefits of these approaches could be utilized. To address this challenge we developed the Throughput Ranking by Iterative Analysis of Genome Enrichment (TRIAGE) approach which iteratively applies enrichment analysis and predicted protein-protein interactions to subset the data from high-throughput screening studies with increased statistical correction of the final result for both false positive and false negative targets. To test the utility of TRIAGE analysis we analyzed the findings from three screens of host dependency factors in HIV previously reported to lead to discordant results. We found that re-analysis of the primary data by TRIAGE measurably improved the concordance across these studies and simultaneously identified shared biological enrichments and novel targets. These findings suggest that applying the existing bioinformatic toolkit through an iterative framework optimizes the hit selection of omics-scale studies by robustly interrogating the range of enrichments and interactions represented in the results. The TRIAGE platform is publicly available as a web-based user interface at triage.niaid.nih.gov




AND
Emrah Gecili, Ph.D
Post-Doctoral Scholar
Cincinnati Children's Hospital Medical Center


Nominated by Rhonda Szczesniak, Ph.D

"Making proteomics-informed predictions of rapid cystic fibrosis disease progression"

Rapid lung disease progression can occur throughout the lifespan in individuals with cystic fibrosis (CF). Although it is critical to intervene prior to irreversible lung damage, there is limited information on predictive biomarkers for this purpose. Numerous models and machine learning methods have been posited for predicting disease progression; however, little has been done to develop these models with proteomic markers for prediction rather than discrimination. We examined the extent to which rapid lung disease progression in CF could be more accurately predicted with inclusion of proteomics samples. Mass spectrometry was performed on serum samples from 88 individuals with CF aged 6-18 years who also had longitudinal information on lung function and other clinical/demographic characteristics. We compared feature selection results through sequential and parallel procedures using LASSO, random forests, principal components regression and marginal testing techniques, in order to identify important features among 5011 protein isoforms while accounting for clinical/demographic characteristics. We found that a robust set of 12 protein isoforms can be identified across this broad range of approaches. We examined the extent to which this set of markers improved prediction of rapid disease progression. The prediction model was based on a Gaussian linear mixed effects model with nonstationary covariance, thereby allowing us to model lung function trajectories while simultaneously incorporating proteomic expression levels and patient covariate histories. The model was also used to calculate real-time risk of rapid lung function decline. An app for these predictions was developed to facilitate use in point-of-care settings. Accuracy was assessed by using 5-fold cross-validation and ROC curves to predict rapid decline. Forecast validation examining the predictive ability of the proteomic model after serum collection showed improved accuracy in coverage probability and ROC analysis for rapid decline. Coverage probability was found by determining whether the observed slope over the two lung function measurements, subsequent to protein measurement, is contained is the 95%CI from the prediction model. For example, a single protein isoform improved coverage probability from 64.3% to 71.4%, enabling better predictive performance. Proteomics-informed prediction modeling more accurately forecasted rapid decline than modeling with clinical/demographic characteristics alone.


Seminars will be webcast online at http://videocast.nih.gov and available on the
Proteomics Interest Group website http://proteome.nih.gov as an archived presentation unless otherwise noted.



NIH Special Interest Groups home page NIH Home Page

This site was updated on October 31st, 2018. Please contact Renee Olano at olanol(at)mail.nih.gov with questions or suggestions.