bioconductor package gene network inference with ensemble of trees (genie3) release 3.12  (RStudio)

Journal: Bioinformatics

Article Title: Fast and accurate gene regulatory network inference by normalized least squares regression

doi: 10.1093/bioinformatics/btac103

Figure Lengend Snippet: Maximum node degree for predicted GRNs of varying SNR on hub-prone data. Ten datasets were used for GRN inference with LSCON, LSCO, LASSO, RidgeCO and Genie3. The prediction was done on simulated data containing infinitesimal values and the network selected was the predicted GRN with a median degree closest to the median degree of the true GRN, from a set of 30 GRNs of varying sparsity generated for each dataset by each method. The simulated data contained 100 ( A ), 300 ( B ), 500 ( C ) or 800 ( D ) genes corresponding to the titles in the figure. The average maximum degree of the true GRNs is shown as a dotted line. The simulated data were generated from GRNs with scale-free topology

Article Snippet: The methods are LSCO, LASSO, ( Tjärnberg et al. , 2013 , ), ridge regression ( Friedman et al. , 2010 ) with cut-off (RidgeCO) and Genie3 ( Huynh-Thu et al. , 2010 ); note that the Matlab version of Genie3 was used here.

Techniques: Generated

Journal: Bioinformatics

Article Title: Fast and accurate gene regulatory network inference by normalized least squares regression

doi: 10.1093/bioinformatics/btac103

Figure Lengend Snippet: Accuracy of GRNs predicted from hub-prone data. LSCON, LSCO, LASSO, RidgeCO and Genie3 were used to predict GRNs from 10 datasets of 100 ( A ), 300 ( B ) and 500 ( C ) genes, as well as from 5 datasets of 800 ( D ) genes, corresponding to the titles in the figure. The simulated data contained small singular values and was generated from GRNs with scale-free topology. The AUPR is plotted over signal to noise ratio (SNR) as the methods are expected to perform better at higher SNR levels

Article Snippet: The methods are LSCO, LASSO, ( Tjärnberg et al. , 2013 , ), ridge regression ( Friedman et al. , 2010 ) with cut-off (RidgeCO) and Genie3 ( Huynh-Thu et al. , 2010 ); note that the Matlab version of Genie3 was used here.

Techniques: Generated

Journal: Bioinformatics

Article Title: Fast and accurate gene regulatory network inference by normalized least squares regression

doi: 10.1093/bioinformatics/btac103

Figure Lengend Snippet: Mean execution time for five GRNI methods. The methods LSCO, LSCON, LASSO, RidgeCO and Genie3 were run on simulated data of varying sizes and their execution time measured in CPU time. Ten datasets at each size were used for determining average runtime for each method. Due to the excessive runtime of Genie3, we could not run it on the 800 gene datasets as the total runtime exceeded 100 wall clock hours

Article Snippet: The methods are LSCO, LASSO, ( Tjärnberg et al. , 2013 , ), ridge regression ( Friedman et al. , 2010 ) with cut-off (RidgeCO) and Genie3 ( Huynh-Thu et al. , 2010 ); note that the Matlab version of Genie3 was used here.

Techniques:

Journal: Scientific Reports

Article Title: KBoost: a new method to infer gene regulatory networks from gene expression data

doi: 10.1038/s41598-021-94919-6

Figure Lengend Snippet: Regression-based Gene Regulatory Network Inference Algorithms. Schematic describing how different regression based GRN inference methods work. These methods are based on different machine learning algorithms. We show six methods based on different machine learning algorithms that differ on the model shape and the way models for different TFs are combined. They were selected because they represent major types of machine learning methods used for GRN reconstruction and because of their high performance in the DREAM 4 and DREAM 5 challenges. PLSNET uses partial least squares and fits a linear model between TFs and targets. TIGRESS uses a linear model with different lasso parameters They both rely on the assumption that the expression of a gene is proportional to the expression levels of the TFs that regulate it. GRNBoost2 and ENNET use boosting to learn different tree models between TFs and targets. GENIE3 also uses tree models, however they iteratively resample different subsets of observations and potential TFs per target and create an ensemble of tree models. Unlike linear models, tree models do not rely on any assumption between the relationship of a TF and a target, however they are not continuous models.

Article Snippet: GENIE3 (Matlab, 2010) , 0.67 , 0.30 , 0.82 , 0.46 , 0.003 , 37.72 , 1.60 , 40.74 , 806.75.

Techniques: Expressing

Journal: Scientific Reports

Article Title: KBoost: a new method to infer gene regulatory networks from gene expression data

doi: 10.1038/s41598-021-94919-6

Figure Lengend Snippet: IRMA dataset benchmark.

Article Snippet: GENIE3 (Matlab, 2010) , 0.67 , 0.30 , 0.82 , 0.46 , 0.003 , 37.72 , 1.60 , 40.74 , 806.75.

Techniques:

Journal: Scientific Reports

Article Title: KBoost: a new method to infer gene regulatory networks from gene expression data

doi: 10.1038/s41598-021-94919-6

Figure Lengend Snippet: DREAM 4 dataset benchmark.

Article Snippet: GENIE3 (Matlab, 2010) , 0.67 , 0.30 , 0.82 , 0.46 , 0.003 , 37.72 , 1.60 , 40.74 , 806.75.

Techniques:

Journal: Scientific Reports

Article Title: KBoost: a new method to infer gene regulatory networks from gene expression data

doi: 10.1038/s41598-021-94919-6

Figure Lengend Snippet: DREAM 5 dataset benchmark.

Article Snippet: GENIE3 (Matlab, 2010) , 0.67 , 0.30 , 0.82 , 0.46 , 0.003 , 37.72 , 1.60 , 40.74 , 806.75.

Techniques:

Journal: Scientific Reports

Article Title: KBoost: a new method to infer gene regulatory networks from gene expression data

doi: 10.1038/s41598-021-94919-6

Figure Lengend Snippet: Results summary of running time and performance.

Article Snippet: GENIE3 (Matlab, 2010) , 0.67 , 0.30 , 0.82 , 0.46 , 0.003 , 37.72 , 1.60 , 40.74 , 806.75.

Techniques: