Genetic Functional Similarity Clustering Using CRISPR-Cas9 Knockout Data

doi:10.21203/rs.3.rs-3013259/v1

Download PDF

Research Article

Genetic Functional Similarity Clustering Using CRISPR-Cas9 Knockout Data

https://doi.org/10.21203/rs.3.rs-3013259/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background:

Machine learning models have become a powerful tool in the field of genetics, allowing scientists to make more accurate predictions about the functions of genes using currently available information. Utilizing both pre-existing annotations from previous studies and multiple genome-wide experimental data would provide us with the potential to construct a more comprehensive model about the functional similarity between genes.

Results:

In this paper, we used knockout phenotype information obtained from CRISPR-cas9 knockout experiments performed under various conditions and using various cells to improve gene functional similarity prediction. We applied Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) and Agglomerative Hierarchical Clustering algorithms to find functionally linked gene groups from knockout data. Subsequent gene enrichment analysis revealed that gene groups defined with knockout data could be associated with specific biological functionality with a high degree of statistical significance. Furthermore, we were able to identify possible functional similarities between an undescribed gene and previously researched genes by using HDBSCAN labels. As a case study, we manually investigated KCNA1/SCN9A pair, which showed highly similar HDBSCAN label profiles, and identified that they were both associated with Oncogene-Induced Senescence (OIS), information that was not found in available databases.

Conclusion:

We found that previously unaddressed functional similarities between genes could be identified from genome-wide CRISPR-Cas9 phenotype datasets. This approach might help to identify novel biomarkers or potential drug targets for diseases with few therapeutic options.

Gene function

Functional similarity

CRISPR-Cas9

HDBSCAN

Gene enrichment

Attaining a comprehensive understanding of gene function is crucial for advancing the field of life sciences, as it has a broad impact on biology. Despite the numerous advances in biotechnology, which have enabled researchers to examine biological systems with unprecedented scale and accuracy [1], the functional characterization of unknown genes still remains a challenge which calls for substantial resources and efforts.

Recent high-throughput techniques have provided high-quality and large-scale gene annotation data, resulting in the development of various methods [2] for predicting gene functional similarity. Predictions of functional similarity between genes could be generated using methods based on functional annotations of genes within Gene Ontology Database [3]. Networks of genes with associated functions could also be employed to predict the function of a gene of interest [4]. In addition, gene function could be predicted by analyzing multiplex gene expression maps while considering the spatial relationship among the sampled locations [5]. Many other biological data types, such as DNA sequences, protein structures, and protein-protein interactions, have been utilized to create models for predicting gene function [6]. All these methods could help in understanding the characteristics of unknown genes with reduced experimental cost, although they often provide insufficient and even irrelevant information, further vexing the researchers.

CRISPR-Cas9 is a highly precise and targeted mechanism for cutting DNA, revolutionizing the field of gene editing [7]. Compared to older methods, CRISPR-based knockout has numerous advantages: for example, CRISPR-Cas9 based knockout could target DNA sequences with greater precision than RNAi techniques [8]. On the other hand, compared to older gene editing technologies such as zinc finger nucleases (ZFNs) or Transcription activator-like effector nucleases (TALEN), CRISPR-Cas9 based knockout is much more cost- and time-efficient [9], making it a more suitable choice for large-scale screening. Now, using CRISPR-Cas9 system, researchers can conduct high-throughput analyses of the causal interactions between genes and external factors, such as drug treatment or viral infection, in a human cell environment [10–12].

With the growth of genome-wide knockout phenotype data in various experimental settings, we found a potential to use it for constructing a model that could predict the functions of genes. To design a model that provides comprehensive information about the biological roles of genes, we utilized two computational approaches: clustering and gene enrichment analysis. Clustering [13] is a type of unsupervised machine learning techniques [14] that groups data points together based on their similarity without using predefined labels. Clustering algorithms are based on the principle that entities with similar characteristics or functions are grouped, whereas entities with different characteristics or functions are separated into different groups. Algorithms such as Hierarchical Density-based Spatial Clustering of Applications with Noise (HDBSCAN) [15] or Agglomerative Hierarchical Clustering [16] are examples which could be used to group genes with similar phenotypic profiles from various experiments or conditions. We applied these algorithms to analyze gene-phenotype data from CRISPR-Cas9 knockout experiments, thereby identifying functional similarities among genes. After obtaining the clustering results, gene enrichment analysis was used to identify a group of genes that displayed a statistically significant relationship to specific biological keywords [17], providing insights into the shared properties among genes within the group.

In this research, we aim to demonstrate how we can predict the functional profiles of genes of interest using CRISPR-Cas9 knockout phenotype profiles. The phenotypic datasets from independent CRISPR-Cas9 knockout experiments performed under various experimental settings were collected and utilized to achieve this goal. Two aforementioned clustering algorithms and gene enrichment analysis were subsequently applied to extract functional similarity profiles from numerical phenotype data. To examine the efficacy of our workflow, a case study utilizing the gene pairs identified through gene pair analysis was performed using 137-label datasets identified through HDBSCAN (Fig. 1).

Clustering-based functionality analyses of genes based on the CRISPR knockout screening data subsets using HDBSCAN

We first designed a study to examine the functional similarity of genes at a more precise level, using CRISPR knockout screening data. On the one hand, analyzing individual screening data independently has several limitations, such as an insufficient data size and inherent biases originating from the experimental settings, consequently limiting model performances. On the other hand, using the entire screening datasets at once may lead to reduced prediction accuracy due to imbalanced sample sizes among the experimental data samples, or the dilution of meaningful signals coming from a small subset of data.

To overcome these shortcomings, we devised an analysis workflow that merges two screening datasets from separate studies into one, followed by a clustering analysis using HDBSCAN. This approach might address multiple aforementioned issues: merging two datasets increases the length of the feature vector assigned for each gene, which partly neutralizes experimental setting-dependent biases and thereby increases the model's robustness. Furthermore, we could generate unique subsets from the available data, establishing a 'population of clustering results' that enables statistical analysis. This may allow us to better examine differences between clusters and similarities among genes with the same label within a cluster.

We were able to analyze 137 unique subsets using this workflow (Supplementary table 2) and identified several notable instances. By clustering data subset generated by merging experimental results of two cell proliferation-related studies, Behan [18] and Tzelepis [19], we found several interesting distribution patterns within t-SNE space. We analyzed 104 genes, marked number 9, which were clustered into one spot (Figure 2A). The Gene Ontology database [20]-based enrichment analysis revealed that 32 out of 104 genes were associated with cytoplasmic translation (Fold Enrichment = 50.60, P-value = 2.38E-43). Additional 11 genes were found to be associated with the U2-type prespliceosome assembly (Fold Enrichment = 89.87, P-value = 1.26E-17) (Figure 2D). In this gene list, we identified FAM86B1 as the only gene with no known function.

Interestingly, we found that CRISPR-Cas9 knockout of the genes associated with label number 9 resulted in very high levels of cell death. To compare with our current knowledge about cell death and survival, we examined the locations of genes that are known to be specifically associated with the cell death pathway within the t-SNE [21] space (Figure 2B). Contrary to our initial assumption, the distribution of cell death-associated genes within t-SNE space did not provide any meaningful information, implying a possible discrepancy between mechanistic association and actual phenotypic involvement.

Similar clustering analysis was done with a data subset generated by merging the experimental results of two studies in different categories, DNA damage response-related screening [22] and cell growth screening [23] (Figure 2C). We examined 1022 genes associated with cluster label number 21. Among the 1022 genes studied, 471 were linked to nucleobase-containing compound metabolic processes, according to the gene ontology database. (Fold Enrichment = 3.31, P-value = 2.04E-125). Moreover, 373 additional genes were associated with the RNA metabolic process (Fold Enrichment = 4.53, P-value = 5.24E-131) (Figure 2E). We also found EBNA1BP2 gene with unknown function.

Functionality analysis of genes based on whole CRISPR data using agglomerative hierarchical clustering

In addition to functionality analyses based on data subsets, we analyzed whole CRISPR knockout data using a hierarchical aggregation clustering algorithm, in order to investigate features as a whole. Genes were assigned to one of the seven clusters (Figure 3A, 3B): we found knockouts of genes in clusters 3, 4, and 6 evidently triggered cell death in almost all cell types (strong red and blue signals in Figure 3B), whereas the patterns were less distinct for the other clusters.

To further visualize and assess the functional characteristics of gene clusters, we used Appyters [24], which projects input gene lists to a pre-computed UMAP space and detects associations with Wikipathway biological pathways [25]. We found that Cluster 1 had the strongest association with three biological processes: adipogenesis, the IL17 signaling pathway, and DNA Damage Response (Figure 3C, F). We also confirmed the presence of C1QTNF4, which is predicted to activate cytokine activity [26], within this cluster. On the other hand, genes within Cluster 2 were grouped in pink and orange clusters on Wikipathway UMAP, indicating the strongest correlation with phosphodiesterases involved in neural function (Figure 3D, G). The presence of the DBNL gene in Cluster 2 supports this result, as it is believed to be involved in activities such as nervous system development, Rac protein signaling, and podosome assembly [27]. Clusters 3, 4, and 6 showed largely similar patterns, and genes in these clusters were largely grouped together in green clusters on UMAP (Figure 3E, H). These clusters showed the strongest connection with the mitochondrial electron transport chain and the oxidative phosphorylation system. Interestingly, we found the AURKAIP1 gene within this cluster, which had previously only been associated with the upstream or positive regulation of proteolysis [28].

Functionality analysis of genes using similarities of data subset-wise clustering patterns

As we mentioned in the previous sections, it is possible to generate a pool of clustering results by using subsets of available data. Based on the assumption that genes with similar functions are likely to be close together in the feature space, we compared the clustering labels (n=137) of each gene and searched for pairs with high label overlap values, to see if these pairs could represent interesting emergent properties.

Using more than 20,000 gene pairs with clustering label overlap values greater than 80% (Supplementary table 5), we generated a functional similarity network which revealed several potential functional clusters (Figure 4). To analyze each cluster of genes and discover distinct annotation patterns, we employed the Enrichr platform [29], which performs gene enrichment analysis using multiple biological databases. We identified 29 genes linked to the dark blue cluster that are associated with cholesterol metabolism and related signaling pathways. Similarly, other gene clusters showed the strongest association with purine metabolism (green, 60 genes), nephrotic syndrome (yellow, 107 genes), vitiligo (pink, 121 genes), and Notch signaling (gray, 55 genes), respectively.

Further investigation of the aforementioned mechanisms revealed a clear association between them and cell death. High cholesterol is known to trigger both apoptosis and autophagy [30] and purine metabolism clearly plays a role in apoptosis [31]. Nephrotic syndrome increases the apoptosis rate of circulating lymphocytes [32], and vitiligo - an autoimmune disease of the skin - involves a gradual loss of melanin cells [33]. Finally, high Notch signaling is clearly associated with altered cellular fate [34]. This clearly demonstrates that cell death is a recurrent theme among all the mechanisms related to the gene list we obtained from our network analysis.

Case studies of gene pairs

We conducted a case study using gene pairs from the analysis that had a high clustering label overlap (Table 1). In the case of the HRH1/CHRM3 pair, which shows 82.48% clustering label overlap, we identified explicit functional associations in all four databases (Gene ontology, Reactome reaction, pathway [35], and KEGG pathways [36]). In the case of the ONECUT1/IRF7 pair, which shows slightly lower 80.29% clustering label overlap, we found a research paper that states these are two of the three genes most closely associated with inflammatory responses related to head and neck cancer [37]. One interesting case is KCNA1/SCN9A gene pair, which shows 82.48% clustering label overlap and known functional commonality as a voltage-gated ion channel. After further examination, we uncovered that these genes are both associated with senescence caused by oncogenes (OIS) [38, 39], which has not been described in available databases.

Table 1 Case study gene pair list

Gene pair name 1	Gene pair name 2	Label duplication probability	Journal name
ONECUT1	IRF7	80.29%	Pilot Study of Combined Aerobic and Resistance Exercise on Fatigue for Patients with Head and Neck Cancer: Inflammatory and Epigenetic Changes
SFRP1	SLIT3	82.5%	Comprehensive DNA Methylation and Mutation Analyses Reveal a Methylation Signature in Colorectal Sessile Serrated Adenomas
KCNA1	HRH1	82.5%	Different genes may be involved in distal and local sensitization: A genome‐wide gene‐based association study and meta‐analysis
HRH1	LDLRAD4	82.5%	1. Histamine Induces Upregulated Expression of Histamine Receptors and Increases Release of Inflammatory Mediators from Microglia 2. Epigenetics of neuroinflammation: Immune response, inflammatory response and cholinergic synaptic involvement evidenced by genome-wide DNA methylation analysis of delirious inpatients
GIMAP4	STAT4	82.5%	Genetics of Behçet's disease lessons learned from genomewide association studies
SOX8	SFRP1	82.5%	RNA-seq reveals downregulated osteochondral genes potentially related to tibia bacterial chondronecrosis with osteomyelitis in broilers
HERC6	SFRP1	82.5%	NA methylation in demyelinated multiple sclerosis hippocampus
AJAP1	SOX8	82.5%	Methylomics analysis identifies epigenetically silenced genes and implies an activation of β-catenin signaling in cervical cancer

* Tables are sorted alphabetically, and when using data, for data with fewer screens (for example, SARS CoV 2 data or Protein/peptide accumulation data), the same phenotypic data are combined into a single data set. If the number of genes used in one experimental paper is different for each screen, it is described as the number of genes.

In this study, we utilized two clustering algorithms and gene enrichment analysis to discover functional similarities among genes by analyzing phenotypic data from CRISPR-Cas9 gene knockout experiments. Instead of only analyzing the results of individual experiments, we incorporated data from multiple experimental conditions, such as viral response, drug response, and cell proliferation, which enabled us to gain insights into novel features that are currently not present in existing biological databases.

By utilizing our workflow, we were able to predict the functions of previously unaddressed genes. We validated our results by identifying gene pairs with a label replication rate of over 80%, and we demonstrated their high quality using independent research data. We found that the overall mechanism in the gene pair network was related to cell death, which might be a natural consequence of using the CRISPR-Cas9 screening system: which implies that our method would be particularly effective for identifying gene pairs with cell death-related functions. This suggests the possibility of utilizing our workflow to identify novel drug targets against cancer, which is intrinsically a disease of cell proliferation and death, especially by focusing on gene clusters that cause cell death in all cells. However, due to the inherent limits of the screening system, which only measures cell death and survival, the data conveyed within the phenotype information is somewhat limited, making it difficult to identify physiological mechanisms which are only distantly associated with cell fate.

To overcome these limitations, further in-depth investigation is necessary, such as CRISPR-Cas9 knockout experiments under a wider range of conditions that have not been previously addressed, or double knockout experiments to find out the effects of specific gene combinations. Implementing other machine-learning techniques could be another option for efficiently extracting embedded functional information, which could augment our reasoning of gene functions from available data.

CRISPR-Cas9 knockout data sources

To identify the genome-wide functional relationships represented by knockout screening data, we utilized BIOGRID-ORCS [40], a specialized repository of CRISPR-Cas9 experimental data. We collected screening results from 37 independent studies, which observed the individual effect of at least 16,974 gene knockouts on one of seven phenotypes [see Additional file 1].

Subset-wise clustering with HDBSCAN

We generated subsets of knockout data by merging the results from two independent studies. Individual results were normalized using MinMaxScaler from Sklearn library [41], and then genes appearing only in one study were dropped. Subsequently, two tables - which now have the same number of genes - were combined, resulting in a single merged table.

Total of 137 subsets were generated as a result [see Additional file 1]. HDBSCAN is used to analyze each merged table and identify clustering patterns: to maximize clustering efficiency, different min_cluster_size (minimum size of cluster) parameter were given for each data subset. We tested eight parameter options (min_cluster_size = 2 ~ 9) individually and selected the one that resulted in the most even clustering results, consequently extracting the clustering labels for each gene. To properly visualize clustering results, dimensionality reduction of the feature space is done with the t-SNE from Sklearn library (Figs. 2A ~ 2C).

Whole dataset-based clustering with Agglomerative hierarchical clustering

Agglomerative hierarchical clustering was used to validate data from various perspectives. We used all available datasets except for those from Behan FM and Meyers RM papers, as the number of individual screens associated with these papers were significantly high (325 and 340 respectively: [see Additional file 1]), which would have caused an imbalanced dataset if included. We standardized the dataset using StandardScaler from Sklearn library: to standardize the extreme values, a value of 3 was designated if the original value was greater than 3 and − 3 if it was less than − 3.

With agglomerative hierarchical clustering, we extracted 7 clusters that had the highest similarity based on CRISPR phenotypic numerical criteria [see Additional file 1]. Cluster hierarchy was visualized as a dendrogram using Scipy library (Fig. 3A & 3B).

Gene ontology enrichment analysis

Gene Ontology enrichment analysis of the gene list identified from subset-wise clustering is done using the tool provided within the Gene Ontology (GO) database. We used options utilizing GO and Reactome annotations to infer the biological context of each gene list and attempted to further validate our analysis results with information from Reactome and KEGG databases, which cover information about biological processes, biomolecular pathways, and pathophysiological associations.

On the other hand, gene ontology enrichment analysis of gene sets identified from whole dataset-based clustering is done with Appyters platform, which utilizes Wikipathway data and allows identification of shared functionality between genes.

Calculation of clustering label overlap percentage

We calculated clustering label overlap percentage (value) using the labels generated from 137 subset-wise analysis [see Additional file 1]. For each sub-analysis, we checked if the clustering labels of two genes were the same: a 'match' result was returned if the clustering labels were the same, and an 'unmatch' result was returned if either the clustering labels were different, or one or more gene was absent from that sub-analysis (Fig. 5). Repeating this process for all 137 analyses generated a feature vector with boolean values for each gene pair, which could be converted into label overlap percentage (1).

$$\frac{\text{Number of overlapping labels between gene A and gene B}}{137}\times 100\%$$

We found gene pairs with at least 60 matched clustering labels (43.8% label overlap) may show statistically significant functional association: to ensure reasonable precision, we selected gene pairs with more than 80% label overlap percentages (109 matched labels) and conducted a case study.

Network Analysis

We generated a connectivity network using 20,333 gene pairs with clustering label overlap percentages of 80% or higher [see Additional file 1]. Individual nodes represent genes involved in one of the gene pairs, and the color of each node indicates the number of associated genes: blue for nodes with more than 300 associated genes, green for those with 190 ~ 300 associated genes, yellow for those with 82 ~ 190 associated genes, pink for those with 10 ~ 82 associated genes, and gray for all others respectively. Each edge was assigned a value which indicates the clustering label overlap percentage between two nodes. Visualization is done using the Kamada Kawai layout network algorithm from NetworkX [42], resulting in Fig. 4.

This study demonstrated the utility of clustering algorithms in predicting and analyzing genes with similar functions using experimental data from CRISPR knockout experiments. We applied two clustering algorithms to gene phenotypes obtained from experiments performed under various experimental conditions, in order to identify functional similarities among genes and discover previously unknown, functionally related gene pairs. The data used in this study have limitations in identifying specific mechanisms: however, the suggested technique would offer more comprehensive and varied results than those obtained from more traditional approaches that use only a small number of experimental conditions. The methods presented in this paper would provide valuable guidance for future studies on the functional similarities among genes.

Ethics approval and consent to participate

Not applicable

Consent for publication

Not applicable

Availability of data and materials

Not applicable

Competing interests

Not applicable

Funding

Not applicable

Competing interests

The authors declare that they have no competing interests.

Author’s contributions

JL analyzed the function prediction by Machine learning and the properties of the functional similarity analysis. JL, HH, KN and KC drafted the paper. MC jointly designed the study and finalized the manuscript. All authors read and approved the final manuscript.

Acknowledgments

We would like to thank Editage (www.editage.co.kr) for English language editing.

Author information

Authors and Affiliations

Department of The Integrative Biotechnology & Translational Medicine, Yonsei University, Incheon 21983, Korea

Ji on lee, Hansol Hong, Kyoung Tai No

Bioinformatics and Molecular Design Research Center (BMDRC), Incheon 21983, Korea

Min Hyung Cho, Kyoung Tai No

School of Systems Biomedical Science and Research Center for Integrative Basic Science, Soongsil University, Seoul 06978, Korea

Kwang Hwi Cho

corresponding author

Correspondence to Kyoung Tai No

Wang Z, Gerstein M, Snyder M: RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet 2009, 10(1):57-63.
Zhao Y, Wang J, Chen J, Zhang X, Guo M, Yu G: A Literature Review of Gene Function Prediction by Modeling Gene Ontology. Front Genet 2020, 11:400.
Teng Z, Guo M, Liu X, Dai Q, Wang C, Xuan P: Measuring gene functional similarity based on group-wise comparison of GO terms. Bioinformatics 2013, 29(11):1424-1432.
Tian Z, Guo M, Wang C, Liu X, Wang S: Refine gene functional similarity network based on interaction networks. BMC Bioinformatics 2017, 18(Suppl 16):550.
An L, Ling H, Obradovic Z, Smith DJ, Megalooikonomou V: Learning pair-wise gene functional similarity by multiplex gene expression maps. BMC Bioinformatics 2012, 13 Suppl 3(Suppl 3):S1.
Smaili FZ, Tian S, Roy A, Alazmi M, Arold ST, Mukherjee S, Hefty PS, Chen W, Gao X: QAUST: Protein Function Prediction Using Structure Similarity, Protein Interaction, and Functional Motifs. Genomics Proteomics Bioinformatics 2021, 19(6):998-1011.
Ran FA, Hsu PD, Wright J, Agarwala V, Scott DA, Zhang F: Genome engineering using the CRISPR-Cas9 system. Nat Protoc 2013, 8(11):2281-2308.
Fire A, Xu S, Montgomery MK, Kostas SA, Driver SE, Mello CC: Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature 1998, 391(6669):806-811.
Gaj T, Gersbach CA, Barbas CF, 3rd: ZFN, TALEN, and CRISPR/Cas-based methods for genome engineering. Trends Biotechnol 2013, 31(7):397-405.
Meyers RM, Bryan JG, McFarland JM, Weir BA, Sizemore AE, Xu H, Dharia NV, Montgomery PG, Cowley GS, Pantel S et al: Computational correction of copy number effect improves specificity of CRISPR-Cas9 essentiality screens in cancer cells. Nat Genet 2017, 49(12):1779-1784.
Schleicher EM, Dhoonmoon A, Jackson LM, Clements KE, Stump CL, Nicolae CM, Moldovan GL: Dual genome-wide CRISPR knockout and CRISPR activation screens identify mechanisms that regulate the resistance to multiple ATR inhibitors. Plos Genet 2020, 16(11).
Ramaker RC, Hardigan AA, Gordon ER, Wright CA, Myers RM, Cooper SJ: Pooled CRISPR screening in pancreatic cancer cells implicates co-repressor complexes as a cause of multiple drug resistance via regulation of epithelial-to-mesenchymal transition. Bmc Cancer 2021, 21(1).
Hartigan JA, Wong MA: Algorithm AS 136: A K-Means Clustering Algorithm. Journal of the Royal Statistical Society Series C (Applied Statistics) 1979, 28(1):100-108.
Barlow HB: Unsupervised learning. Neural computation 1989, 1(3):295-311.
McInnes L, Healy J, Astels S: hdbscan: Hierarchical density based clustering. J Open Source Softw 2017, 2(11):205.
Day WHE, Edelsbrunner H: Efficient Algorithms for Agglomerative Hierarchical-Clustering Methods. Journal of Classification 1984, 1(1):7-24.
Reimand J, Isserlin R, Voisin V, Kucera M, Tannus-Lopes C, Rostamianfar A, Wadi L, Meyer M, Wong J, Xu C et al: Pathway enrichment analysis and visualization of omics data using g:Profiler, GSEA, Cytoscape and EnrichmentMap. Nat Protoc 2019, 14(2):482-517.
Behan FM, Iorio F, Picco G, Gonçalves E, Beaver CM, Migliardi G, Santos R, Rao Y, Sassi F, Pinnelli M et al: Prioritization of cancer therapeutic targets using CRISPR–Cas9 screens. Nature 2019, 568(7753):511-516.
Tzelepis K, Koike-Yusa H, De Braekeleer E, Li Y, Metzakopian E, Dovey OM, Mupo A, Grinkevich V, Li M, Mazan M et al: A CRISPR Dropout Screen Identifies Genetic Vulnerabilities and Therapeutic Targets in Acute Myeloid Leukemia. Cell Rep 2016, 17(4):1193-1205.
Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, Foulger R, Eilbeck K, Lewis S, Marshall B, Mungall C et al: The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res 2004, 32(Database issue):D258-261.
Van der Maaten L, Hinton G: Visualizing data using t-SNE. Journal of machine learning research 2008, 9(11).
Olivieri M, Cho T, Alvarez-Quilon A, Li K, Schellenberg MJ, Zimmermann M, Hustedt N, Rossi SE, Adam S, Melo H et al: A Genetic Map of the Response to DNA Damage in Human Cells. Cell 2020, 182(2):481-496 e421.
Chai AWY, Yee PS, Price S, Yee SM, Lee HM, Tiong VK, Goncalves E, Behan FM, Bateson J, Gilbert J et al: Genome-wide CRISPR screens of oral squamous cell carcinoma reveal fitness genes in the Hippo pathway. Elife 2020, 9.
Clarke DJB, Jeon M, Stein DJ, Moiseyev N, Kropiwnicki E, Dai C, Xie ZR, Wojciechowicz ML, Litz S, Hom J et al: Appyters: Turning Jupyter Notebooks into data-driven web apps. Patterns 2021, 2(3).
Martens M, Ammar A, Riutta A, Waagmeester A, Slenter DN, Hanspers K, R AM, Digles D, Lopes EN, Ehrhart F et al: WikiPathways: connecting communities. Nucleic Acids Res 2021, 49(D1):D613-D621.
Li Q, Wang L, Tan W, Peng Z, Luo Y, Zhang Y, Zhang G, Na D, Jin P, Shi T: Identification of C1qTNF-related protein 4 as a potential cytokine that stimulates the STAT3 and NF-κB pathways and promotes cell survival in human cancer cells. Cancer letters 2011, 308(2):203-214.
Boateng LR, Cortesio CL, Huttenlocher A: Src-mediated phosphorylation of mammalian Abp1 (DBNL) regulates podosome rosette formation in transformed fibroblasts. J Cell Sci 2012, 125(Pt 5):1329-1341.
Lim SK, Gopalan G: Aurora-A kinase interacting protein 1 (AURKAIP1) promotes Aurora-A degradation through an alternative ubiquitin-independent pathway. Biochem J 2007, 403:119-127.
Kuleshov MV, Jones MR, Rouillard AD, Fernandez NF, Duan Q, Wang Z, Koplev S, Jenkins SL, Jagodnik KM, Lachmann A et al: Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 2016, 44(W1):W90-97.
Li KQ, Deng Y, Deng GM, Chen PY, Wang YT, Wu HT, Ji ZG, Yao ZL, Zhang XR, Yu B et al: High cholesterol induces apoptosis and autophagy through the ROS-activated AKT/FOXO1 pathway in tendon-derived stem cells. Stem Cell Res Ther 2020, 11(1).
Camici M, Garcia-Gil M, Pesi R, Allegrini S, Tozzi MG: Purine-Metabolising Enzymes and Apoptosis in Cancer. Cancers (Basel) 2019, 11(9).
Zachwieja J, Bobkowski W, Zaniew M, Dobrowolska-Zachwieja A, Lewandowska-Stachowiak M, Siwinska A: Apoptosis and antioxidant defense in the nephrotic syndrome. Pediatr Nephrol 2003, 18(11):1116-1121.
Ruiz-Arguelles A, Brito GJ, Reyes-Izquierdo P, Perez-Romano B, Sanchez-Sosa S: Apoptosis of melanocytes in vitiligo results from antibody penetration. J Autoimmun 2007, 29(4):281-286.
Yang XD, Klein R, Tian XL, Cheng HT, Kopan R, Shen J: Notch activation induces apoptosis in neural progenitor cells through a p53-dependent pathway. Dev Biol 2004, 269(1):81-94.
Vastrik I, D'Eustachio P, Schmidt E, Gopinath G, Croft D, de Bono B, Gillespie M, Jassal B, Lewis S, Matthews L et al: Reactome: a knowledge base of biologic pathways and processes. Genome Biol 2007, 8(3):R39.
Kanehisa M, Furumichi M, Tanabe M, Sato Y, Morishima K: KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res 2017, 45(D1):D353-D361.
Xiao C, Beitler JJ, Higgins KA, Chico CE, Withycombe JS, Zhu Y, Zhao H, Lin IH, Li F, Jeon S et al: Pilot study of combined aerobic and resistance exercise on fatigue for patients with head and neck cancer: Inflammatory and epigenetic changes. Brain Behav Immun 2020, 88:184-192.
Lallet-Daher H, Wiel C, Gitenay D, Navaratnam N, Augert A, Le Calve B, Verbeke S, Carling D, Aubert S, Vindrieux D et al: Potassium channel KCNA1 modulates oncogene-induced senescence and transformation. Cancer Res 2013, 73(16):5253-5265.
Warnier M, Flaman JM, Chouabe C, Wiel C, Gras B, Griveau A, Blanc E, Foy JP, Mathot P, Saintigny P: The SCN 9A channel and plasma membrane depolarization promote cellular senescence through Rb pathway. Aging Cell 2018, 17(3):e12736.
Oughtred R, Rust J, Chang C, Breitkreutz BJ, Stark C, Willems A, Boucher L, Leung G, Kolas N, Zhang F et al. The BioGRID database: A comprehensive biomedical resource of curated protein, genetic, and chemical interactions. Protein Sci. 2021, 30(1):187-200.
Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: Machine learning in Python. Journal of machine learning research. 2011, 12(Oct):2825–30.
Hagberg A, Swart P, S Chult D: Exploring network structure, dynamics, and function using NetworkX. In.: Los Alamos National Lab.(LANL), Los Alamos, NM (United States); 2008.

No competing interests reported.

Additionalfile1.xlsx
Additional file 1 Contains a list of data used in the analysis and four other result data. This file contains a list of data used for analysis. Furthermore, it includes a sheet with a list of label numbers assigned to each gene, obtained by using the specified minimum number of clusters and the entire dataset during HDBSCAN execution. Additionally, after performing agglomerative hierarchical clustering, each gene is showed a specified label number, and a separate sheet provides a list of gene pairs that exhibit over 80% label duplication, suitable for gene pair analysis. (.xlsx)

Download PDF

Version 1

posted

You are reading this latest preprint version

Genetic Functional Similarity Clustering Using CRISPR-Cas9 Knockout Data

Status:

Version 1

Abstract

Figures

Background

Results

Discussion

Methods

CRISPR-Cas9 knockout data sources

Subset-wise clustering with HDBSCAN

Whole dataset-based clustering with Agglomerative hierarchical clustering

Gene ontology enrichment analysis

Calculation of clustering label overlap percentage

Network Analysis

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1