Recursive partitioning tools have become prevailing and widely used methods for nonparametric regression and classification in varieties of scientific fields. Random forests, which can deal with a large number of predictor variables even in the existence of complex interactions, have been successfully applied in genetics, epidemiology, and clinical medicine within the past few years. As a well-recognized expert in the virtual screening field, Creative Biolabs has introduced multiple ligand-based screening services, including recursive partitioning methods, to advance their projects for rationally designed drug candidates.
Recursive partitioning (also known as classification and regression trees) is a more recent technique that can provide a visualization of the main structure of a regression relationship. It’s a prediction method frequently used with dichotomous outcomes to avoid the assumptions of linearity. The tool involves constructing a decision tree by repeatedly dividing a data set into subsets based on descriptors, or rules, that discriminate between different types of molecules. And each subdivision is formed by splitting the sample on the value of specific predictor variables. The means (e.g., cross-validation) of fitting the tree to data, help reduce overfitting (including unnecessary predictor variables), especially in cases with many potential predictors. Furthermore, it needs to be highlighted that recursive partitioning method is exploratory rather than inferential technology.
Up to now, recursive partitioning has gained popularity as a means of multivariate data exploration in the analysis of microarray data, DNA sequencing, and many other applications in genetics, medicine, and bioinformatics. It is an essentially fairly simple nonparametric technique for prediction and classification. When used in a standard way, the tool frames trees showing the succession of rules that require to be followed to receive a predicted value or class.
Recursive partitioning, as a simple visual display, explains the appeal to researchers from different disciplines. At Creative Biolabs, we’d like to introduce the principles of recursive partitioning techniques as well as current methodological improvements to clarify their usage for low and/or high dimensional data exploration. The strategy is a data-mining method using statistical tests to identify descriptors of objects that split one class from another. Also, this technique takes advantage of molecular descriptors to discover those that separate actives from inactives.
Our expert teams design recursive partitioning protocols that work in a tree fashion. One variable is used to make the first split, then all molecules in each subgroup are divided independently, one descriptor at a time, until either compound is completely classified or further splits won’t meet the statistical significance criterion. A recursive partitioning analysis leads to a tree in which every branch is the value of a certain property and every end node is enriched in active or inactive compounds.
Notably, one of the most important methods is named the “random forest” approach. A random forest is a set or ensemble of classification or regression trees. Each tree in this ensemble is built according to the mechanisms of recursive partitioning, where the feature space is recursively divided into regions with observations of similar response values.
Recursive partitioning remains an area of active research. It’s a statistical method extremely appropriate for detecting ligand-based drug discovery in virtual screening. As a top-ranking expert in the biologics market, Creative Biolabs provides sophisticated recursive partitioning techniques to create a decision tree that strives to correctly classify members of the sample by splitting it into subgroups based on several dichotomous independent variables. Our services could assist clients to identify active compounds from a giant database and promote their development on small molecular screening. If you’re interested in any hit identification steps, please don’t hesitate to contact us.
For Research Use Only.
