NLM IRP Seminar Schedule

Multiomics Data Integration using Graph Convolutional Networks

To pave the road towards precision medicine in cancer, patients with similar biology ought to be grouped into same cancer subtypes. Utilizing high-dimensional multiomics datasets, integrative approaches have been developed to uncover cancer subtypes. Recently, Graph Neural Networks have been discovered to learn node embeddings utilizing node features and associations on graph-structured data. Some integrative prediction tools have been developed leveraging these advances on multiple networks with some limitations.

In this talk, a new method called SUPREME is introduced. SUPREME is a node classification framework, which integrates multiple data modalities on graph-structured data. On breast cancer subtyping, unlike existing tools, SUPREME generates patient embeddings from multiple similarity networks utilizing multiomics features and integrates them with raw features to capture complementary signals. On breast cancer subtype prediction tasks from three datasets, SUPREME outperformed other tools. SUPREME-inferred subtypes had significant survival differences, mostly having more significance than ground truth, and outperformed nine other approaches. These results suggest that with proper multiomics data utilization, SUPREME could demystify undiscovered characteristics in cancer subtypes that cause significant survival differences and could improve ground truth label, which depends mainly on one datatype. In addition, to show model-agnostic property of SUPREME, we applied it to two additional datasets and had a clear outperformance.

Identification of fold-switching proteins by FLIM-FRET

Historically, a protein’s sequence has been thought to provide information for only one fold. Recent work has not only identified that proteins can adopt two distinct folds with different functions but that this phenomenon, called fold switching, occurs in nature frequently. The current limit in understanding and observing these fold-switching proteins is that expressing them is often complicated with solubility issues, and protein structure prediction software has bias based on the assumption proteins only have one possible fold. One such fold-switcher RfaH, a protien in the NusG/Spt5 family, assumes an ⍺-helical hairpin fold capable of autoinhibition that limits specifity of opsDNA binding and reduces off-target competition with NusG in E. coli. RfaH adopts a β- roll fold that can directly interact with the S10 integral ribosomal subunit and increase translation through arresting pausing of the ribosome. The β -roll fold of RfaH is most similar to it’s parent protein NusG which also functions to continue translation by pausing the arrest of the ribosome. The similarity in function between RfaH and NusG is caused by similarity in the β -roll fold and sequence of the N-terminal domain (NTD), whereas RfaH differs in it’s unique ability to fold-switch to its ⍺-helical hairpin fold by difference in it’s C-terminal domain (CTD).

Techniques commonly used for characterizing two folded states are Circular Dichroism (CD), NMR, x-ray crystallography, and cryo-EM, all of which are laborious, expensive, and require a lot of pure protein. However, recent developments in confocal microscopy have enabled new higher-throughput assays to be developed. Here we present an assay capable of distinguishing between the ⍺-helical hairpin of RfaH and β- roll of NusG by Förster resonance energy transfer (FRET) efficiency determined by fluorescent lifetime intensity microscopy (FLIM), also termed FLIM-FRET.

We hypothesize the end-to-end distances between RfaH and NusG can be measured by FLIM-FRET using GFP at the NTD and mCherry at the CTD. Our preliminary data indicates there is often a quantifiable and consistent FRET efficiency difference between RfaH and NusG variants. Ongoing work includes using this new higher throughput tool to investigate the evolution of RfaH in the NusG/Srt5 family and, in conjunction with a new computational tool, to identify potential fold-switchers.

TBD