Varied Signature Made Easy

Varied signature

hi my name is lucas goldman and today i will be talking about my project l 1000 kd squared analysis and visualization of the lynx l1000 shrna gene knockdown dataset i completed this research as a summer scholar in the mayan lab just some quick background in this project gene expression is measured through the l1000 assay how this works is that there are these luminex beads that can be coated with rna probes for a specific gene the color of the bead indicates which gene it corresponds with when the bead is washed with a cell line's rna it becomes bright based on the gene's expression level the l1000 is limited to use 500 different bead colors but a technique allows each color to correspond with two genes as the l1000 is able to measure 1 000 expression levels so it can measure 978 genes and 22 controls these genes called the landmark genes were specifically chosen because they maximize the number of genes that could be accurately predicted based on their expression levels so a model takes in the expression levels for the 978 genes and infers the expression level of 11 350 genes thus after the full process 12 328 gene expression levels are known and a pro of the l1000 is that 384 wells can be measured together using a fully automated process at the broad institute if one applies a perturbation meaning a disease drug g knockdown or anything that alters the state of a cell line and they measure that cell line's gene expression one can assess what genes are upregulated and down regulated in that cell line and this is what is referred to as a signature for that perturbation lynx or the library of integrated network based cellular signatures is an nih-funded research initiative a goal of links is to develop a vast library of these cellular signatures my project is creating a tool to enable users to explore a data set from the lynx library the last thing to review is short hairpin or shrna gene knockdown this is the process of silencing the expression of a specific gene how this works is that the shrna is introduced to the cell through a plasmid and forms a protein called rna-induced silencing complex through this protein whenever a strand of mrna of the target gene is produced that mrna is cleaved and completely silenced the problem with shrna is that while it does knock down the target gene it is also known to have off-target effects this is combated by performing the knockdown in triplicate and using three distinct shrnas of the same knockdown gene where this fits into my project is that for each signature analyzed in this project a cell line had a specific gene knocked down then an l1000 assay was performed to measure the gene expression why this is interesting is that you can learn a lot about the role of a gene by examining how a cell line changes after this gene has been knocked down in this project 119 848 unique gene knockdown signatures were examined these signatures varied in three ways first there were a total of 4 371 different genes knocked down in this data set and each signature had one of these genes knocked down each signature is from one of 15 cell lines and lastly signatures varied based on how long after gene knockdown the l1000 assay was performed all this data is publicly available on geo under the listed accession number this data is in the form of gene expression levels and was transformed into differential gene expression signatures using the characteristic direction method which computes relative gene expression and allows one to determine which genes are up and down regulated in a signature each signature contains differential gene expression levels for each gene thus it has 12 328 dimensions we can visualize this data in 2d by reducing its dimensionality through pca t-sne and umap each of these are shown using the project's data as you can see the umap was the best in clustering similar gene signatures which is valuable for recognizing patterns in the data and is why umap was what was eventually used in this project this data was initially around 120 000 signatures when plotting each signature as you can see in the leftmost image there are simply too many data points to see any real clustering and it is very dense in addition not all genes are expressed in every cell and knocking down a gene in a cell line that it's not expressed in does not affect that cell thus the data had to be filtered to remove the signatures not representative of their knockdown gene the qualification used to determine this was whether the signature was correlated above a certain value with at least one signature of the same knockdown gene the graphic shows the umap visualization at different cutoff scores and eventually a cut-off score of 0.25 was decided on as it removed enough signatures to allow visible clustering to occur but still kept 2991 genes the react scatter board developed by this lab was used to display the visualization to demonstrate the use of this visualization using the covid19 geneset library also developed by this lab i looked at the genes that are frequently in coven-19 gene sets and found that stat1 appeared the most next as you can see on the right i searched the umap visualization for stat 1 and the yellow circles are where signatures of that gene can be located next i zoomed into one of the yellow clusters to see what other genes are similar to this gene that is highly associated with covid we see this adcy9 gene that appears four times in this cluster which may suggest a potential relationship between these two genes adcy9 is adenylate cyclase 9 and it codes for a membrane-bound enzyme that catalyzes the formation of cyclic amp from atp when adcy9 is searched in arches 4 for its predicted biological process some of the top hits are related to negative regulation of t cells which are significant in the autoimmune response to covet 19. adcy9's connection to coven 19 has not been explored and this example shows an interesting use of this aperture in finding and exploring relationships between genes the aperture also allows the user to submit a signature to be projected on this data set each user is shown the following screen and are asked to input the name of a disease or small molecule that they would like to compare to this data set after this a signature of this input is fetched from creeds the user also has the option to input their own signature the example input is a signature of sars cov2 from the human lung cell on this is the signature that will be used as a case study for the remainder of this presentation after the users click submit similarity scores are calculated between each knockdown signature and the input signature similarity is equal to the overlap minus the opposite overlap divided by the size of the input gene set where overlap means the number of up and down genes shared between two signatures and opposite overlap is the number of genes that are up in one signature and down in the other along with this z scores are calculated using the distribution of similarity scores over all signatures and p-values are calculated using the fischer's exact test the user is given this list of the most similar and opposite genes and next the aperture allows the user to see what they have in common to do so the aperture submits a gene set of the knockdown genes in the 50 most similar and opposite signatures to enricher enricher allows the user to see transcription factors pathways ontologies and more that are relevant to the submitted gene set i have taken two of the enricher results from these gene sets the knockdown genes in these similar signatures overlap with the gene set of severe combined immunodeficiency and what this is showing is that the same genes that when knocked down mimic covid are also related to immunodeficiency quick reminder of what opposite signatures are genes that are highly expressed in opposite signatures are down regulated in the sars kovi's two signature and vice versa and richer results show that the knockdown genes of these signatures are involved in chronic inflammatory diseases to conclude the l1000 kd squared aperture is accessible it allows anyone to analyze genes even if they lack a background in bioinformatics l1000 kd squared can be used to discover relationships between human genes based on their effects on cells when knocked down l1000kd squared can be used to identify genes that play important roles in the pathophysiology of diseases and the mechanisms of drugs by submitting drugs and disease signatures as queries to the aperture in the future more benchmarking must be done for this after to make sure that the results are as accurate as possible it would also be interesting to visualize and analyze other l1000 datasets in the same way this includes drug and small molecule data and crispr knockout data which would be interesting to compare to the knock down data in this project lastly i'd like to incorporate the l1000fwd into this aperture which is a visualization and analysis tool for small molecules this would allow the user to identify pathways affected by drugs i would like to thank the entire mayan lab for their help specifically sherry thank you so much for setting up and running this amazing program my collaborators dr lachman and daniel clark thank you for answering my constant stream of questions thank you to all the summer scholars including jessie who supplied the stars kobe 2 signature that i presented and charles and skyler who helped me create fields for my aperture and avi thank you so much for giving me constant support and guidance thank you for listening