ADMET includes drug absorption, distribution, metabolism, excretion and toxicity. One main reason for drug R&D failures is the efficacy and safety deficiencies which are related largely to ADMET properties. In the early study of ADMET properties, functional proteins from human or humanized tissues have been usually used as drug targets. Common research technologies combined with computer simulation are applied to study the interaction between drugs and targets, revealing biophysical and biochemical influences in vivo. Prediction of ADMET is an important process in drug design and drug screening. However, the current experimental methods for ADMET evaluation are still costly and time-consuming. In order to further improve the accuracy of ADMET property prediction, based on Big Data, Machine Learning and Deep Learning technologies, CD ComputaBio is dedicated to effectively extracting structural features (including processing small molecule and protein structure) through deep neural network algorithm. It virtually predicts and evaluates ADMET and other properties of small molecule structures on cell, protein and disease level. Function modules in the platform enable our clients to conveniently perform several types of drug-likeness analysis, ADMET endpoints prediction, systematic evaluation and database/similarity searching.
All the obtained data are collected from peer-reviewed publications, public databases through manually filtering and processing.
Pretreatments are carried out to guarantee the quality and reliability of the data. Remove compounds that without explicit description for ADME/T properties Reserve only one entity if there are duplicate or more same compounds. Reduce the random error when fluctuations of compounds values in a reasonable limit Wash molecules by MOE (disconnecting groups/metals in simple salts, keeping the largest molecular fragment and add explicit hydrogen).
Molecular descriptors, which include constitution, topology, connectivity, E-state, Kappa, basak, burden, autocorrelation, charge, property, MOE-type descriptors, etc., and fingerprints are applied to further model building.
Eliminate some uninformative and interferential descriptors Build regression models using fivefold cross-validation method
Different modeling algorithms are applied to develop regression or classification models for ADME/T related properties, including RF, SVM, RP, PLS, NB and DT.
To evaluate various ADMET properties, a series of high-quality prediction models would be generated and validated, applying fivefold cross-validation and using commonly used parameters.
Tech Trends in Artificial Intelligence What's up, what's down, what lies ahead, only found in MedAI!