Enalos KNIME nodes

Blending cheminformatics, bioinformatics and nanoinformatics tools under KNIME

Categories:

Design of Experiments nodes

Box Behnken

Ιn statistics, Box–Behnken designs are experimental designs for response surface methodology to achieve the following goals: a. Each factor is placed at one of three equally spaced values. b. The design should be sufficient to fit a quadratic model. c. The ratio of the number of experimental points to the number of coefficients in the quadratic model must be in reasonable limits i.e. in the range of 1.5 to 2.6.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Central Composite

A Box-Wilson Central Composite Design is commonly called “central composite design”, or response-surface-methodology (RSM) contains an embedded factorial or fractional factorial design with center points that is augmented with a group of ‘star points’ that allow estimation of curvature.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Fractional Factorial

A fractional factorial design is a reduced version of the full factorial design, meaning only a fraction of the runs are used. A fractional factorial design allows for a more efficient use of resources as it reduces the sample size of a test, but it comes with a tradeoff in information i.e. the factors must be confounded.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Full Factorial

A full factorial experiment is an experiment whose design consists of two or more factors, each with discrete possible values or "levels", and whose experimental units take on all possible combinations of these levels across all such factors.

Input A table containing all the factors that are needed to form all possible combinations

Output A table containing all possible combinations of potential experiments

Halton Design

Halton design is based on spreading out points with the low discrepancy Halton sequence.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

LHS

Latin hypercube designs are model independent, space filling designs often used in computer experiments. In these designs each of the k factors is divided into n equal levels such that there is only one run containing a given level of a factor. The number of total runs is also equal to n.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

LHS Space Filling

The space-filling designs, spreads out points with the aim of encouraging a diversity of data once responses are observed. A spread of training examples will ultimately yield fitted models which smooth/interpolate/extrapolate best, leading to more accurate predictions at out-of-sample testing locations.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Plackett Burman

Plackett-Burman experimental design is used to identify the most important factors early in the experimentation phase when complete knowledge about the system is usually unavailable. In practical use, two-level full or fractional factorial designs, and Plackett-Burman designs are often used to screen for the important factors that influence process output measures or product quality. Interactions between the factors are considered negligible. The underlying concept is to find an experimental design where each combination of levels for any pair of factors appears the same number of times, throughout all the experimental runs. A complete factorial design would satisfy this criterion, but Plackett-Burman tries to find smaller designs.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Sukharev Grid

Special property of this grid is that points are not placed on the boundaries of the hypercube, but at centroids of the sub-cells constituted by individual samples. This design offers optimal results for the covering radius regarding distances based on the max-norm.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Uniform

The uniform experimental design is one kind of space filling designs that can be used for computer experiments and also for industrial experiments when the underlying model is unknown. The uniform design seeks its design points to be uniformly scattered on the experimental domain.

Input A table containing all the factors that are needed to form all possible combinations

Output Output table containing the experimental design

Modelling nodes

Consensus

In case that combination of classifiers is desired in a classification scheme, voting provides a baseline method for joining the results of classifiers. With the consensus node, it is possible to combine the predictions of different classifiers on the same data and acquire the majority class of the predictions for each sample. There is no limitation on the number of the combined classifiers, however an odd number of classifiers must be used in order to avoid dissension.

Input Classification results of different classifiers (one column per classifier) for a set of instances

Output Majority vote results (consensus table)

Domain-APD

Applicability Domain (area of reliable predictions) based on the Euclidean distances. The applicability domain must be defined to flag compounds/samples for which predictions of a developed model may be unreliable. In this node similarity measurements are used to define the domain of applicability of the model based on the Euclidean distances among all training compounds and the test or virtual screening compounds. A threshold value is calculated and a prediction is considered reliable only if distances between train and test/screening compounds are lower than this threshold (Zhang et al. 2006).

Input Training set table and test/virtual screening set table (only the descriptors involved in modelling)

Output A table containing for each compound of the test/virtual screening set, the result "reliable"/"unreliable"

Domain-Leverage

Applicability Domain (area of reliable predictions) based on the extent of extrapolation. Extent of extrapolation is one simple approach to define the applicability domain and it is based on the the diagonal elements of the Hat matrix (leverage values h). These values reflect the similarity of test/screening samples to the training set (distance from the training set’s centroid) based on the descriptor values used in the model development. The limits of the applicability domain are determined by a threshold leverage value h*. The prediction for a test sample is considered reliable if h is lower than h* (Golbraikh and Tropsha 2020).

Input Training set table and test/virtual screening set table (only the descriptors involved in modelling)

Output A table containing for each compound of the test/virtual screening set, the result "reliable"/"unreliable"

EnalosDeNovo

With EnalosDeNovo node the user is enabled to create a list of molecules by combining a series of substituents with a core molecule. All the possible combinations of molecules can be then used as inputs to the nodes Valence Checker (in order to check the feasibility of molecules), Molecule to CDK (to observe the structure) etc.

Input A table containing the substituents of the base molecule

Output A table with all possible combinations of substituted molecules

EnaloskNN

The k-nearest neighbors algorithm (kNN) is a type of instance-based learning (lazy) method used for classification and regression. In this method, the prediction for the unknown endpoint of an instance, is calculated by taking into consideration its k closest training neighbors in the feature space. The prediction is the weighted average (in regression) or the weighted majority vote (in classification) of the endpoints of the k nearest neighbors. The neighbors have different contributions to the prediction depending on their distance from the test instance and this is achieved using as weighting factors the inverted distance values.

Input Training data and test data

Output Predicted endpoint values for each test instance and k neighbors of the training set

Int2Double

Int2Double node converts all (or a subset of) the integer columns of an input table, to double value columns.

Input A table and a list of the integer columns to be converted to double

Output The table with converted columns

Kennard and Stone

The Kennard-Stone method allows the selection of two representative subsets (e.g., training and test sets for modelling purposes) with a uniform distribution over an initial dataset. Data samples selection is based on their Euclidean distances in the data space (Kennard and Stone, 1969 and Daszykowski, Walczak and Massart, 2002).

Input Table to be split

Output Two separate tables containing the partitioned data

MLR

MLR node performs Multiple Linear Regression in order to model the relationships between a scalar dependent variable y and two or more independent variables denoted as X.

Input The train and test data in tabular format

Output A table containing the coefficient of determination (R squared), constant (y-intercept) and the regression coefficients for each independent variable of the multiple linear model, as well as the experimental and the predicted test values of the dependent variable y

Model Acceptability Criteria

This node gives information about the quality of fit and predictive ability of a continuous QSAR model based on the criteria proposed by Golbraikh and Tropsha, 2002.

Input A table containing for the test set the predicted (ypred) and the actual values (yexp) of the dependent variable, and a table containing the actual dependent variable values for the training set (ytr).

Output The quality of fit and predictive ability statistics of a continuous QSAR-type model along with an indication for the fulfilment of the relevant criteria.

PLS

PLS node, performs Partial Least Squares (PLS) regression analysis by applying the SIMPLS algorithm.

Input The train and test data in tabular format

Output A table containing the constant and a matrix, containing intercept terms (beta), as well as the experimental and the predicted values of the dependent variable y

PLSLoadings

PLSLoadings node, performs the calculation of the loadings on the given data by applying the SIMPLS algorithm.

Input The training set data in tabular format

Output A table with the X and Y loadings

PLSscores

PLSScores node, performs the calculation of the scores on the given data by applying the SIMPLS algorithm.

Input The training set data in tabular format

Output A table with the X and Y scores

Remove Column

Remove Column node removes the input table columns that contain the same values at a percentage equal or higher than a specified cutoff limit.

Input A table with arbitrary input data

Output The filtered table

Remove Duplicates

Remove Duplicates enables the user to remove the rows of the input table that contain the same values in selected columns. The filtered table contains all rows that are unique and the first one of each repeated row.

Input Table with arbitrary input data

Output Filtered table

Sphere Exclusion

Sphere Exclusion node allows the selection of two representative subsets (such as training and test sets) using an exclusion radius and attempts to specify compounds which most effectively cover the available data space. The minimum exclusion radius is selected as the desired percentage of the maximum Euclidean distance from the center. All compounds with a distance from the center lower than the exclusion radius are assigned to the test set and are removed from the initial dataset. The remaining compound which is closest to the center is assigned to the training set and is removed from the initial dataset, as well. At each iteration, a new center is selected from the samples included in the train set (Hudson, et al. 1996).

Input Table to be split

Output Two separate tables containing the partitioned data

Y-randomization

Y-randomization (or Y-scrambling) is a technique, applied to ensure a QSAR model’s robustness. This test consists of repeating modelling with scrambled values of the response variable (endpoint) of the training set. If a model built with randomized endpoints of the training set has statistically significant lower predictive power on the training or on the test set, then the model built with the original endpoints is reliable. Otherwise, an acceptable QSAR model cannot be obtained for the specific modeling method and data (possible overfitting or chance correlation).

Input Endpoint data and number of randomizations to be performed

Output Randomized endpoint columns

Molecular descriptors nodes

Mold2

Mold2 calculates a large and diverse set of molecular descriptors (777) encoding two-dimensional chemical structure information. The software is developed by the Center for Bioinformatics at the National Center for Toxicological Research (NCTR).

Input A licensed installation of Mold2 file (exe file) and the compound structures as an SDF chemical file

Output A table containing for each compound the values of 777 molecular descriptors calculated with Mold2

CDK

CDK (Chemistry Development Kit) is a java based descriptor calculation tool which calculating topological, geometrical, charge based and constitutional descriptors.

Input The compound structures as an SDF chemical file

Output A table containing for each compound the values of 287 CDK molecular descriptors

PubChem nodes

Assay

Assay node gives the user access to PubChem database via substance or compound IDs (SID and CID), in order to find the Assays where a particular compound is tested. Using this node the user can download in KNIME information about the Assay and the Assay outcome.

Input Table with at least one column with the following structure identifiers: InChIKey, PubChem CID, IUPAC name and SMILES for a compound.

Output Extracted information from PubChem: Assay name, AID (BioAssay identification number), Bioassay Type, Bioactivity Outcome, Target IG, Target GeneID, Activity Value in μM (whenever available) and Activity Name.

Assay Class

Assay Class node searches in PubChem database according to one or more given AIDs (BioAssay identification numbers) and displays only the active or inactive compounds.

Input Table with one column containing the AIDs

Output Table with extracted information from PubChem: PubChem SIDs (Substance IDs) of compounds that are active or inactive

Main PubChem

Main PubChem node enables the user to search the PubChem database and obtain the following information for thousands of compounds with one request: PubChem CID (Compound ID), IUPAC Name, InChI, InChI-Key Molecular Formula, Molecular Weight, Canonical SMILES and the direct PubChem URL.

Input Table with at least one column from the following structure identifiers: InChIKey, PubChem CID, Name and SMILES (each column can contains thousands of structures)

Output Extracted information from PubChem: PubChem CID, IUPAC Name, InChI, InChIKey Molecular Formula, Molecular Weight, Canonical SMILES, PubChem URL etc.

Patent

Patent node gives the user straight access to the PubChem database in order to obtain information about the patent coverage information for thousands of compounds with one request.

Input Table with at least one column from the following structure identifiers: InChIKey, PubChem CID, Name and SMILES (each column can contains thousands of structures)

Output Extracted information from PubChem: Patent name

Patent to Sid

Patent to Sid node helps the user to search the PubChem database and obtain the SIDs (Substance IDs) of the compounds covered by the patents in request.

Input Table with one column with the Patent IDs

Output Table with extracted information from PubChem: PubChem SIDs (Substance IDs) of compounds that are covered by the patents in request. In case that no compounds are found under a patent, the number -1 is assigned as SID of this patent

Sid

Sid node exports the CIDs (Compound IDs) of a given list of SIDs (Substance IDs), searching the PubChem database. The user can search the PubChem database and obtain information about the CIDs for thousands of compounds with one request.

Input Table with one column containing the SIDs

Output Extracted information from PubChem: PubChem CID and synonyms

Similarity

Via Similarity node, the user can search the whole PubChem database for similar compounds (Tanimoto Similarity) and obtain the following information for thousands of compounds with one request: PubChem CID (Compound ID), Molecular Formula, Molecular Weight and Number of Rotatable Bonds.

Input Table with at least one column from the following structure identifiers: InChIKey, PubChem CID, Name and SMILES (each column can contains thousands of structures)

Output Extracted information from PubChem: PubChem CID, Molecular Formula, Molecular Weight and Number of Rotatable Bonds

Vendor

Vendor node enables the user to search the PubChem database and obtain information about the commercial availability for thousands of compounds with one request.

Input Table with at least one column from the following structure identifiers: InChIKey, PubChem CID, Name and SMILES (each column can contains thousands of structures)

Output Extracted information from PubChem: PubChem SID, Compound URL, Supplier, Category of the Vendor

UniChem nodes

UniChem

Enalos UniChem nodes gives to the user direct access to UniChem databases (e.g., ChEMBL,PubChem, DrugBank, KEGG, ZINC, BindinDB, eMolecules, IBM patents, SureChEMBL etc.) through KNIME. UniChem is a superset of all 27 available databases, separated in 5 friendly and easily recognizable categories.

Input Table with at least one column with InChiKey

Output The number of databases were the input compound are found, along with information about these databases, the compound IDs and links for the compounds in these databases

UniChem Connectivity

UniChem Connectivity is an expanded version of the standard UniChem tool that allows you to find related molecules. Connectivity Search allows molecules to be first matched on the basis of complete identity between the connectivity layer of their corresponding Standard InChIs, and the remaining layers then compared to highlight stereo-chemical and isotopic differences.

Input Table with at least one column with InChiKey and search criteria

Output The number of databases were the input compound are found, along with information about these databases, the compound IDs and links for the compounds in these databases

Contact NovaMechanics Ltd

info[at]novamechanics[dot]com

NovaMechanics Ltd

NovaMechanics Ltd

Powered by w3.css

Web Analytics