PTML: Perturbation-Theory Machine Learning notes

Abstract

PTML: Perturbation-Theory Machine Learning methods have been developed by H. to seek models able to predict multiple properties f(si, cj)k of type k of a system (si) at the same time (multi-output and multiobjective) taking into consideration variations (perturbations) in multiple experimental conditions cj = (c0, c1, c2, ..... cn) at the same time with respect to a value of reference or expected. PTML-like models have been applied for different authors to study drugs, proteins, nanoparticles, complex networks, social systems, etc. [1-17]. In the particular case of a PTML linear models we can fit an equation with the general form f(si, cj)new = a0 + a1• f(si, cj)ref + SUM(bk•PTO(cj)k). In this model PTO(cj)k are PT operators measuring the perturbations in the new system si with respect to the system of reference sr with observed or expected property f(si, cj)ref. First, we need to calculate the values of the PTOs in the data pre-processing step. This PTOs allow us to perform an Information Fusion process with variables and conditions from different sources. Moving Averages (MA), Multi-condition MA (MMAs), Double MAs, Co-variance Operators, etc. are some examples of useful PTOs. After that, we can use Multiple Linear Regression (MLR), Linear Discriminant Analysis (LDA), or other linear ML techniques to seek the PTML model. In the nonlinear cases, we can fit the PTML models using Artificial Neural Networks (ANN), Support Vector Machines (SVM), Classification Trees, and other ML methods.

Implementation of IFPTML Computational Models in Drug Discovery Against Flaviviridae Family

“…Each feature is assigned a score that represents its importance. The higher the score, the more influential the feature is in making predictions . In tree-based models such as gradient boosting models, the importance is calculated based on the number of times a feature is used to split the data across all trees.…”

“…The goal was to find the best model-based on three statistics parameters to satisfy. Since the models must have statistical values higher than 75%, 36 a balance between specificity, sensitivity, and accuracy was sought. In consequence, the model will be able to correctly predict compounds that have high probabilities of being active and differentiate between active and inactive compounds.…”

“…The higher the score, the more influential the feature is in making predictions. 36 In tree-based models such as gradient boosting models, the importance is calculated based on the number of times a feature is used to split the data across all trees. However, its importance is not a definitive measure of causality.…”

“…They are based on a reference property f ( v ij ) ref . Perturbation operators PT are added to this property to measure the deviations that can occur with respect to this reference property; in this way, it can predict the properties of an unknown system but similar to the original system that was used as a refs – .…”

Perturbation Theory Machine Learning Modeling of Immunotoxicity for Drugs Targeting Inflammatory Cytokines and Study of the Antimicrobial G1 Using Cytometric Bead Arrays

“…The perturbation theory machine learning (PTML) modeling technique is useful to seek predictive models for complex data sets with multiple big data features. , We can predict scoring function values f ( v ij ) calc for the i th compound in the j th preclinical assay with multiple conditions of assay c j = ( c 0 , c 1 , c 2 , ..., c n ). PT operators are used as input.…”